PERSPECTIVES

The adult cortical language network

OPINION
A description of the developing language
system requires a description of the adult
The ontogeny of the cortical language system. Any neurocognitive
model of language comprehension
language network must be seen not only as an inventory
of cognitive capacities that have to be
acquired by a developing speaker, but also
Michael A. Skeide and Angela D. Friederici as an inventory of the neuronal circuits
that have to mature in their specific and
Abstract | Language-processing functions follow heterogeneous developmental fully efficient functionality. The present
trajectories. The human embryo can already distinguish vowels in utero, but description focuses on areas that have been
grammatical complexity is usually not fully mastered until at least 7 years of age. repeatedly implicated as being functionally
Examining the current literature, we propose that the ontogeny of the cortical specialized for language using fMRI, NIRS,
language network can be roughly subdivided into two main developmental stages. MEG and EEG. Domain-general areas
(for example, underlying working memory
In the first stage extending over the first 3 years of life, the infant rapidly acquires or cognitive control) that contribute
bottom‑up processing capacities, which are primarily implemented bilaterally in to language comprehension are not
the temporal cortices. In the second stage continuing into adolescence, top-down considered here (for a review, see REF. 13).
processes emerge gradually with the increasing functional selectivity and An overview of the current findings on
structural connectivity of the left inferior frontal cortex. the neural basis of language production
can be found in BOX 1.

Language is a unique evolutionary children have paved the way to start Bottom‑up comprehension processes.
achievement considered to have decisively systematically investigating how the neural Language comprehension as a process
driven the sophistication of human bases of specific language aspects emerge8–12 unfolds in time. Initially, it is driven by
cognition1. Comprehending spoken (see Supplementary information S1 (box)). bottom‑up processes that begin early
language, which is the focus of this article, The knowledge gathered in these studies, (with onsets between 20 and 120 ms after
is the first challenge children have to which examine the relationships between the utterance is heard), are completed
master on their way towards becoming language behaviour, brain function and rapidly (with durations of 30–60 ms), are
proficient language users, and it remains brain structure, provides the basis for a unconscious and run fully automatically.
a fundamental cognitive ability throughout first attempt to outline the ontogeny of the At the earliest stage of bottom‑up
life. The complexity of this ability lies cortical language network. processing, the acoustic-phonological
in the fact that it requires the brain to In this article, we introduce a neural features of the single sound segments
process multiple heterogeneous types of blueprint of language acquisition to explain (phonemes) of each word are determined.
information. A well-orchestrated interplay where, when and how the maturing brain Such features are extracted in the bilateral
of several cortical regions is needed to masters the language-processing steps that superior temporal sulcus (STS) and, if
segment the incoming auditory stream into are finally carried out by the mature brain. the first syllable of a word provides an
words that can be associated with meaning We describe the processing capacities that unambiguous cue for rapidly reconstructing
(semantics) and combined into sentences represent the milestones to be reached the rest of the word, phonological word
following certain rules (syntax). in the course of ontogeny, and propose that forms can be detected within a time window
Findings from electroencephalography the development of these capacities takes of 20–50 ms14–16. Phonological word-form
(EEG), magnetoencephalography (MEG) place during two developmental stages: information undergoes morphosyntactic
and particularly functional MRI (fMRI) a first one that mainly — but not exclusively categorization — that is, the assigning of
experiments in adults have recently made it — involves data-driven, bottom‑up heard word forms to a syntactic group (such
possible to formulate novel neural models processes, and a second one that involves as noun, verb or preposition) — between 40
of language comprehension that now cover advanced, top-down processes. ‘Bottom‑up and 90 ms after the utterance17. In addition,
the entire processing cascade from audition processing’ here describes the low-level lexical–semantic categorization — which
to the interpretation of an utterance2–4. computation of mental representations is necessary to determine whether a heard
Neural accounts of the developing language from the sensory input, whereas ‘top-down word form can be semantically interpreted
system are so far largely based on EEG processing’ denotes the further analysis — occurs between 50 and 80 ms in the left
data5–7; however, in recent years, advances of these representations under the anterior STS and superior temporal gyrus
in adapting near-infrared spectroscopy influence of more-complex, higher-order (STG)18,19. Lexical–semantic categorization
(NIRS), MEG and fMRI settings for young mental representations. is necessary to identify, for example,

NATURE REVIEWS | NEUROSCIENCE VOLUME 17 | MAY 2016 | 323

©
2
0
1
6
M
a
c
m
i
l
l
a
n
P
u
b
l
i
s
h
e
r
s
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
PERSPECTIVES

Box 1 | Language production in Broca’s area of sentence-level syntax is driven by

modulation of the left posterior superior
In 1865, Pierre Paul Broca, a French physician studying patients with brain lesions, suggested that temporal cortex by the left IFG35,36.
the left inferior frontal cortex has a crucial role in producing language104. This view is still widely Syntactic information is transferred
accepted to this day, and Brodmann area 44 (BA44) and BA45 of the left inferior frontal gyrus are dorsally from BA44 along the arcuate
collectively named ‘Broca’s area’ after its discoverer.
fasciculus (AF) to the posterior STG and
To initiate the action of speaking, a speaker has to select, access and retrieve a particular entry in
the mental lexicon105. The duration of the trajectory from the intention to speak to the selection of STS, and semantic information is transferred
a lexical item cannot be measured directly, but lexical selection triggered by a visually presented ventrally from BA45 and BA47 along the
word takes about 200 ms within the anterior BA45 (REF. 106). This was demonstrated in patients IFOF to the left posterior STG2,24,27,32,37–39.
with epilepsy who, during presurgical preparation, had local field potentials recorded from depth Note that the dorsal connection between
electrodes implanted in Broca’s area while they carried out a language production task. In this BA44 and the left posterior STG and STS that
task, participants were asked to produce the plural form of the word ‘horse’, requiring them is involved in syntactic processes40 can be
to retrieve the stem ‘horse’ and the plural suffix ‘s’ to produce ‘horses’. At around 320 ms after differentiated from a second dorsally located
visual presentation of the word stem, a central portion of BA45 is recruited to extract the fibre bundle that connects the left posterior
morphosyntactic features that are needed to inflect it (that is, to add the ‘s’). Then, phonological STG and STS to the premotor cortex and is
features of the word stem and all added inflectional morphemes are encoded in a posterior part of
involved in the repetition of speech38.
BA45 by approximately 450 ms after visual presentation106. Finally, BA44 and BA45 of the inferior
frontal gyrus together regulate the conversion of the mental representation of a phonological Prosodic information — that is,
word form into an articulation code that can be propagated to the motor cortex to perform the intonation, tone, stress and rhythm
actual articulation. information — is processed in a broad time
Importantly, although Broca’s area has classically been implicated in speech production, it is not window of 200–600 ms after the utterance,
directly involved in the motor act of speaking but instead coordinates the transformation of predominantly in the right STG and the
phonological to articulatory information. Broca’s area starts activating when a visually presented right IFG32,41,42. After at least 600 ms, all
word that has to be repeated is still shown on a screen, but as soon as overt speech articulation information is assumed to be integrated into
commences, it deactivates, while activity in the motor cortex remains107. Finally, given its relatively a complex conceptual representation in the
late activation onset (around 200 ms after visual presentation of a word), Broca’s area does not left posterior STG and STS, with a potential
appear to be directly involved in primary phonological access (which begins at about 20 ms) or
involvement of the left IFG2,3,23,32 (FIG. 1).
lexical access (which begins at about 50–80 ms). Rather, Broca’s area seems to regulate the further
processing of lexical information and phonological inputs that it receives from the temporal cortex Cross-domain interactions — for example,
via long-distance white-matter fibre tracts2,106. between prosody and syntax or between
semantics and syntax — that occur before
the final integration stage are not described
pseudowords (that is, meaningless sound are consciously controllable and do not run here (for reviews, see REFS 2,3).
sequences resembling the acoustic features entirely automatically. Between 200 and
of a word) as being uninterpretable. 400 ms, lexical information is delivered from Developing bottom‑up processes
Once an interpretable word is identified, the left anterior STG and STS along the In this section, we describe the ontological
the corresponding lexicon entry — a mental ventrally located inferior fronto-­occipital emergence of bottom‑up language
representation associated with a phonological fasciculus (IFOF) to Brodmann area 45 processing in approximately the first
word form, carrying semantic and word (BA45) and BA47 of the left inferior 3 years of life. The skills acquired during
category information — can be accessed frontal gyrus (IFG; also known as Broca’s this developmental phase enable infants to
and lexical items can be mentally ‘retrieved’ area), where semantic ambiguity and segment the speech input into phonological
between 110 and 170 ms. This process semantic relations between lexical items are word forms. On the basis of this ability,
is associated with neural activity in the determined through interaction with the they can associate phonological word forms
left anterior STG and STS, and several inferior parietal cortex 21,24–28. Between 300 with semantic representations stored in
other regions widely distributed over the and 500 ms, morphosyntactic information is memory (that is, lexicon entries) that can
cortex 18,20,21. In parallel, as soon as syntactic transmitted ventrally from the left anterior be semantically categorized. Moreover, the
categories are identified, phrase structures STG and STS to the frontal opercular cortex child acquires the ability to assign words to
can be built in the left anterior STG and STS, and BA44, which support the rule-guided syntactic categories and to group words into
possibly together with the most ventral part construction of phrases and sentences29. phrase structures. Although the first years
of the inferior frontal cortex, between 120 It is assumed that this information is of life are dominated by the acquisition of
and 150 ms after the utterance17,22,23. As an transferred along the uncinate fasciculus, bottom‑up language-processing mechanisms,
example, recognizing the determiner “the” but this hypothesis remains to be verified it is important to note that there is also
and the noun “boy” as elements of the phrase experimentally. BA44 also supports the evidence for the first manifestations of
“the boy” requires phrase structure building. ordering of phrases and, together with top-down-based integration of syntactic and
the left posterior STG and STS, enables the semantic information occurring from the
Top-down comprehension processes. building up of sentence-level argument second year of life onwards.
Higher-level comprehension of language structure — that is, verb–noun arrays
is characterized by top-down processes, created in accordance with certain schemes Detection of phonological word forms.
which occur relatively late (with onsets to produce a meaningful sentence24,30–34. Hearing is the first sense to be fully
usually between 200 and 600 ms following Effective fMRI connectivity analyses developed before birth. Accordingly,
the utterance) and proceed relatively modelling the causal interactions between embryos gather their first experiences of
slowly (with durations of at least 150 ms) brain areas during task performance have speech in utero43. A recent study examining
compared with lower-level comprehension, also revealed that top-down processing preterm infants with NIRS — a technique

324 | MAY 2016 | VOLUME 17 www.nature.com/nrn

©
2
0
1
6
M
a
c
m
i
l
l
a
n
P
u
b
l
i
s
h
e
r
s
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
 PERSPECTIVES

a Language comprehension in adults

Bottom-up processes
1 Phonological word form detection
2 Morphosyntactic categorization
3 Lexical–semantic categorization

BA44 4 Lexical access and retrieval

BA45 8 5 Phrase structure reconstruction
7 8 BA44
pSTG 6b
BA41/42 pSTG 6a 6a Prosodic processing
7 FOP pSTS
BA47 1
1 Top-down processes
5 aSTG 1 6b Prosodic processing
4
aSTS 7 Analysis of semantic relations
3
2 8 Analysis of syntactic relations

b
1
2
3
4
5
6a
6b
7
8
0 100 200 300 400 500
Time (ms)

Figure 1 | The adult auditory language comprehension network. areas17,22,23 (stage 5). Phrase-level prosody — that is,Reviews
Nature intonation, tone, stress
| Neuroscience
a | Grey-matter regions in the left perisylvian cortex (indicated by circles, and rhythm information — is processed in a time window of 200–400 ms
shadows and black labelling) and their respective functional roles during after the utterance in the right posterior STG (pSTG)32 (stage 6a). Sentence-
eight central stages of auditory language comprehension (numbered 1–8). level prosody is processed in a time window of 300–500 ms after the utter-
Arrows indicate the assumed flow of information along interconnecting ance in the right inferior frontal gyrus (IFG)41,42 (stage 6b). Between 200 and
white-matter fibre tracts. Bottom‑up processes are coloured in red, and top- 400 ms, lexical information is transferred from the left aSTG and left aSTS
down processes are coloured in blue. b | Time windows (in milliseconds) along the ventrally located inferior fronto-occipital fasciculus (IFOF) to
during which the processing phases that underlie auditory language com- Brodmann area 45 (BA45) and BA47 of the left IFG, where semantic relations
prehension unfold. All values are approximate and subject to inter-­ between lexical items are determined21,24–28 (stage 7). At 300–500 ms after
individual variation. First, phonological word forms are detected within the utterance, phrase structure information is transmitted ventrally from the
20–50 ms after the utterance by extracting acoustic features from the audi- left aSTG and left aSTS, along the uncinate fasciculus (UF), to the frontal
tory cortex14–16 (stage 1). Subsequently, phonological word forms are cat­ opercular cortex (FOP) and to BA44 in the left IFG 29 (stage 8). BA44 is
egorized at the morphosyntactic level between 40 and 90 ms17 (stage 2), and involved not only in ordering phrases but also in building up argument
at the lexical–semantic level between 50 and 80 ms both in the left anterior structures together with the left pSTG and left pSTS, which are connected
superior temporal sulcus (aSTS) and left anterior superior temporal gyrus to BA44 via the left arcuate fasciculus (AF)24,30–34. As soon as an utterance has
(aSTG)18,19 (stage 3). Lexical items associated with phonological word forms gone through these processing steps, the left pSTG and left pSTS, poten-
are retrieved at around 110–170 ms by the left aSTG and left aSTS18,20,21 tially together with the left IFG, can start integrating all information into an
(stage 4). In parallel, phrase structures are built at 120–150 ms in the same interpretable conceptual representation2,3,23,32.

in which 700–900 nm-wavelength light is maturational step is only completed after infants are not more advanced than those
used to measure activity-related changes in week 32 of gestation. It is unlikely that early of age-matched full-term infants46,47. These
blood oxygenation — found that preterm phonological skills are solely the result findings support the view that the primary,
infants born as early as after 28–33 weeks of prenatal environmental experience, as intrauterine speech-perception skills are
of gestation are able to detect a deviant sounds at frequencies above 300 Hz are likely to be driven by genetic factors that
phoneme in a sequence of otherwise strongly attenuated in utero. Given this bias the auditory processing system towards
identical phonemes (for example, ‘ga’ in a level of acoustic degradation, fetuses in the language-specific frequency spectra of the
series of ‘ba’ sounds). Specifically, the NIRS womb can at best only discriminate between acoustic input. Nevertheless, active exposure
technique revealed ‘mismatch’ responses vowels that have lower frequency boundaries to auditory stimuli is crucial for normal
in the bilateral posterior superior temporal (around 200 Hz) but not between more language acquisition, as it facilitates the
cortex and the inferior frontal cortex subtly differing consonants, which in speech differentiation of the sounds in the child’s
in response to the deviant phoneme12. typically have higher frequencies, starting at language environment 48–51.
Remarkably, this basal phoneme- around 300 Hz44,45. Further studies suggest When listening passively to speech,
discrimination capacity is already present that even though preterm infants experience (full-term) newborn babies show strong
when most neurons are still located in the an earlier and richer exposure to speech interhemispheric synchronization
fetal subplate and have not yet migrated that is no longer degraded by maternal of haemodynamic activity in the
into their cortical target layers, as this tissue, speech perception skills in preterm bilateral posterior STG, but no such

NATURE REVIEWS | NEUROSCIENCE VOLUME 17 | MAY 2016 | 325

©
2
0
1
6
M
a
c
m
i
l
l
a
n
P
u
b
l
i
s
h
e
r
s
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
PERSPECTIVES

intrahemispheric synchronization of learning is considered to be crucial for development. This hypothesis should be
haemodynamic activity in the left posterior the ability to memorize relations between tested in follow‑up experiments involving
STG and the left IFG8. However, by the non-neighbouring phonetic elements. younger children.
age of 3 months, infants passively listening Importantly, when newborn babies hear In sum, language acquisition starts
to speech show activity bilaterally in the a sentence, more activity is observed in the in utero, when the embryo gathers first
mid and posterior STG during the first right mid STG than in the left mid STG8. experiences of acoustic information. Infants
3 seconds after stimulus onset, and in Given that, in adults, the right mid STG is learn to discriminate the phonemes of their
inferior frontal regions between 7 and known to support prosodic processing 41,42, target language in the first half-year of life.
9 seconds52. These data are taken to suggest this finding indicates that newborn From 6 months of age onwards, they are
that the interaction between the inferior babies not only are sensitive to segmental able to use prosodic information to segment
frontal motor system and the perceptual (phonological) information but also are speech into phonological word forms.
system functions before, and is required for, equally, or perhaps even more, sensitive These skills are assumed to be based on the
the infant’s babbling stage: a phase in the to suprasegmental (prosodic) acoustic interplay of the bilateral superior temporal
first year of language acquisition in which information in a sentence8. Interestingly, auditory system and the inferior frontal
infants produce articulate speech sounds, a similar lateralization effect was observed motor system.
but not yet comprehensible words, through in the temporo-parietal cortex of sleeping
tuning in towards the phonology of their 3‑month-old infants60. Lexical access and semantic categorization.
target language. The sensitivity of infants to prosody is The refinement of speech segmentation
These findings from the newborn not only vitally important for oxytocin-­ skills in the first 6 months of life provides
babies8 and from the 3‑month-old infants52 supported parental bonding 61 but also the basis for associating phonological word
can be reconciled when considering the forms the basis for learning how to segment forms with objects in the environment, an
structural connectivity pattern in adults, the auditory stream into words, according ability that has indeed been documented
using diffusion-weighted MRI tractography. to the specifics of the target language62. in infants as young as 6 months71. By the
Specifically, adults exhibit two dorsal Event-related potential (ERP) work reveals age of 9 months, infants can generalize
fibre tracts that start in the posterior that, at 6 months of age, infants only the meanings of single words into lexical
temporal cortex: one that projects to the recognize words that have previously been categories according to the visual features
premotor cortex, and one that projects accentuated (that is, marked as acoustically of the objects that they describe, such as
to BA44, which is part of Broca’s area8,53. prominent by increasing sound pressure their colour and shape. At this age, lexical–
In newborn babies8 and 2–5‑month-old and frequency) but not if they have merely semantic learning is dependent on the
infants53, the dorsal fibre tract to the been repeated. It is only at 12 months of age consolidation of recent episodic memory
premotor cortex is already myelinated, but that infants no longer rely on accentuation traces during sleep after the awake encoding
the one reaching into Broca’s area is yet to be to detect phonological word forms63. of experiences72. Support for this view
myelinated. The pathway that projects to the Interestingly, this trajectory coincides comes from adult fMRI experiments that
premotor cortex supports the integration with the decrease in the infant’s ability to suggest that the retrieval of lexical items that
of sensory and motor representations, as discriminate phonemes that do not belong were acquired early in life is controlled by
specifically evident, for example, during to the standard inventory of their native the precuneus (a domain-general episodic
the babbling phase8,53. Moreover, the target language, as indicated by different memory area), whereas the retrieval of
information fed back from the premotor patterns of neural oscillations64–67. The later-learned lexical items is controlled
cortex to the temporal cortex may be crucial declining flexibility in recognizing phonemes by BA45 and BA47 (that is, specialized
for establishing phoneme representations. is known as an instance of perceptual language areas)73. However, an adult-like
This is suggested by the observation that narrowing 68. For example, whereas before semantic processing-associated N400 EEG
6‑month-old infants cannot distinguish the age of 6 months, English-learning response can be elicited at ~12 months
unfamiliar phonemes from one another infants can distinguish between not only in early talkers or ~18 months in normal
when they are prevented from moving their English phonemes but also Hindi phonemes, talkers both in the EEG74 and the MEG11
tongue tips to articulate these phonemes54. they lose the ability to discriminate Hindi in a lexical–semantic priming task (in
Similar work in adults corroborates the phonemes in the second half-year of life49. which word recognition is facilitated if
view that this auditory–motor link is not Crucially, losing this ability coincides with an a target word is preceded by a congruous
only important for language production increased sensitivity for acoustic differences picture but not by an incongruous picture).
but also for language comprehension55. between phonemes of the native language69. The lexical–semantic processing abilities
One could also hypothesize that the dorsal As described above, in the mature brain, required to master this task are purely
network connecting the posterior temporal suprasegmental prosodic information associative, and thus arguably reflect only
and the premotor cortex underlies the is processed at the sentence level in the bottom‑up mechanisms. Importantly,
remarkable ability of infants to implicitly right STG and the right IFG. A similar source localization analyses reveal that the
learn relationships between non-adjacent pattern of dissociable haemodynamic MEG signal corresponding to the N400
syllables56–59. These ‘non-adjacent responses in the right hemisphere (relating in this task originates mainly from mid
dependencies’ usually follow a so‑called to prosodic information) and in the left and posterior portions of the left temporal
AXB structure (for example, ‘le ro bu’), superior temporal cortex (relating to cortex 11, similar to the pattern found in the
in which, according to a predefined rule, segmental phonological processing) has adult brain28.
syllable A (‘le’) predicts that syllable been observed in 4‑year-old children70. Over the course of the second year
B (‘bu’) will follow the interspersed It is likely, however, that this level of of life, infants seem to recognize words
element X (‘ro’)56. This type of rule specialization is reached earlier during not only in a bottom‑up manner but also

326 | MAY 2016 | VOLUME 17 www.nature.com/nrn

©
2
0
1
6
M
a
c
m
i
l
l
a
n
P
u
b
l
i
s
h
e
r
s
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
 PERSPECTIVES

in a top-down fashion. By 14 months and phrase structure (that is, a prosodic Analysis of semantic relations. The
of age, toddlers are already able to boundary occurring before all obligatory emergence of adult-like semantic processing
categorize nouns syntactically based on syntactic elements had occurred) elicited at the sentence level occurs over a long
their knowledge that nouns are preceded a mismatch response in healthy adults but developmental phase. ERP effects associated
by determiners: such toddlers spent less not in patients with lesions in the posterior with the evaluation of semantic congruency
time listening to a phrase containing corpus callosum82. at the word level have an adult-like
a determiner and a pseudoword than Syntactic categorization ability can appearance from 14 months of age onwards,
to a phrase containing a pronoun in be observed by the age of 18 months: and can be considered as prerequisites for
combination with the same pseudoword75,76. 18‑month-old toddlers can already evaluate semantic processing at the sentence level87,88.
This effect was present regardless of which the compatibility of grammatical gender N400 ERP components relating to the
of the two stimulus types the children were information between a case-marked detection of sentence-level incongruency
familiarized with beforehand. At 18 months article and a noun83. For example, French between a verb and a noun (for example, in
old, learning new words requires toddlers detected that the case-marked the phrase “the cat drinks the ball”) can be
simultaneous evaluation of phonological article la (which indexes the feminine detected in children before they are 2 years
features and conceptual-semantic and gender), but not the case-marked article old, although, compared with the adult
syntactic categories. Indeed, 18‑month-old le (which indexes the masculine gender), N400, these components have a later onset
toddlers looked significantly longer at novel was compatible with the feminine noun and last longer 89. As described above, in the
picture–word pairs if the words were from poussette (‘pram’). Two-year-old children mature brain, the semantic and syntactic
different syntactic and semantic categories showed a P600 EEG response (associated domains are dissociable both in terms of
from those of a phonologically similar word with the recognition of syntactic anomalies) their neuroanatomical localization and in
they had known before77. approximately 600 ms after stimulus onset terms of the time windows during which
when they encountered two grammatically their corresponding functions operate.
Syntactic categorization and reconstruction. incompatible phrase types, such as a They only interact at the final stage of
There is behavioural evidence that prepositional and a verb phrase, in German84. sentence comprehension when both types
6‑month-old infants are already able to In addition, 32‑month-old children showed of information have to be integrated into an
recognize phrases if the speech input an early left anterior negativity (ELAN) interpretable conceptual whole; this occurs in
carries reliable prosodic cues. They detect signal that originated from left anterior the left posterior STG and STS, and possibly
noun phrases (for example, “new watches temporal sources17, similar to that observed also the left IFG. A recent fMRI study has
for men”) and verb phrases (for example, during automatic reconstruction of phrases revealed that this modular architecture of the
“buy the whole supply”) if the syntactic in adults85. syntax–semantics interface is implemented
phrase boundary coincides with the It is clear that, at least by the age of in the maturing brain between 7 and 9 years
prosodic phrase boundary, which is marked 3 years, and often before, children are of age. Before this (ages 3–7 years), the
by lengthening of the final syllable and able to process sentences and take part in haemodynamic activity associated with
the pause that succeeds it 78. ERP work verbal communication. Within the present semantics and syntax largely overlaps90,91.
has shown that, by the age of 5 months, model, we assume that they do so by mainly Moreover, there is a statistical interaction
infants are able to ‘chunk’ speech input into involving the temporal cortex and its ventral between the extent of processing of semantics
phonological phrases if the pause (a salient connection to the IFG. This connection has and syntax in mid and posterior portions of
acoustic cue) is available, but not if the pause been shown not only to support semantic the left STG, such that semantic plausibility
is absent 79. An ERP component called the processes at the sentence level38 but also reduces the haemodynamic resources
closure positive shift, which is associated to support the processing of syntactically needed to comprehend syntactically
with the detection of the phonological clause simple sentences37,86 in adults. MRI studies complex sentences. This was demonstrated
boundary, only emerges in the third year involving children below 3 years of age in an experiment that presented either a
of life80. By 6 years of age, children have are needed to explore how the dorsal and semantically plausible proposition (in this
acquired more advanced syntactic phrase ventral connections between the temporal case, a tall animal was the agent performing
structure knowledge, enabling them to cortex and the IFG contribute to language a certain action on a small animal — for
recognize phrases in sentences without the comprehension in the first 3 years of life. example, “Where is the small beetle,
salient pause cues81. whom the big fox carries?”) or, inversely,
One can hypothesize that an inter­ Developing top-down processes a semantically implausible proposition
hemispheric network comprising the The refinement of basic language skills (for example, “Where is the big fox, whom
left anterior STS and STG and the right and increases in working memory the small beetle carries?”). The processing
STG, which are connected by posterior resources together drive the emergence of the non-canonical but plausible sentences
transcallosal fibres, underlies 6‑month-old of higher-order semantic and syntactic consumed less haemodynamic resources
children’s skills in the prosody-enhanced representations. These representations than did the non-canonical and implausible
detection of phrase structure. Low- are specifically related to features relevant sentences; therefore, it is likely that young
frequency-fluctuation MRI analysis has for combined phrases (that is, sentences). children use their conceptual semantic world
revealed interhemispheric functional Top-down processing of semantic knowledge to master syntactic complexity.
connectivity between the respective temporal plausibility and phrase order is given By contrast, 9–10‑year-old children are no
regions in newborn babies8. Support for this special emphasis in the following section. longer biased towards conceptual semantic
functional role of the corpus callosum in Further top-down processes are currently world knowledge to constrain syntactic
adults comes from a study showing that an only studied in adults (for a review, processing. This is reflected at the neural
incongruency between prosodic intonation see REF. 2). level: haemodynamic activity indicating

NATURE REVIEWS | NEUROSCIENCE VOLUME 17 | MAY 2016 | 327

©
2
0
1
6
M
a
c
m
i
l
l
a
n
P
u
b
l
i
s
h
e
r
s
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
PERSPECTIVES

a 3–4 years 6–7 years 9–10 years Adults exhibited no functional selectivity in BA44
or the left posterior STG for syntax. By
contrast, although children aged 9–10 years,
unlike the adults, did not show any such
functional selectivity in the left BA44, they
did show some functional selectivity for
b syntactic complexity in the left posterior
STG: the haemodynamic response of
these children and the adults in this region
was markedly higher for non-canonical
sentences than for canonical sentences.
Remarkably, although 9–10‑year-old
children understand complex non-canonical
Semantic processing Syntactic processing Interaction sentences with an accuracy of more than
Figure 2 | The developmental segregation of syntax from semantics. a | Functional MRI (fMRI) 90%, adults still considerably outperform
responses in children and adults in a sentence comprehension task (thresholded at P < 0.01; them with an accuracy of almost 100%,
Bonferroni- and cluster size-corrected). The haemodynamicNature Reviews
patterns Neuroscience
have a|similar appearance across suggesting that the left BA44 has a decisive
all age groups. Notably, unlike newborn babies and infants8,52,92 but like older children and adults, role in reaching full efficiency in processing
3–4‑year-old children recruit not only the left temporal cortex but also the left inferior frontal cortex syntax 10. This view is corroborated by
when processing sentences (indicated by the arrow). Accordingly, it can be assumed that the brain has structural morphometric data of the brains
set the frame for top-down language processing at 3 years of age. However, the full cortical selectivity of 5–8‑year-old children and adults that
for syntactic and semantic information emerges only gradually within this network. The increasing indicate that increasing syntax skills are
amount of activity in the left inferior parietal cortex could be related to literacy acquisition108. b | The inversely correlated with the grey-matter
development of cortical selectivity for syntactic and semantic information was followed in an fMRI
volume of the left IFG and the left posterior
experiment that used a sentence–picture matching task to investigate the functional effects of syn-
tactic complexity and semantic plausibility. The effects of syntactic complexity were assessed by com- STG (relative to the total grey-matter
paring responses to sentences with non-canonical word order with sentences with a canonical word volume)95. Accordingly, a maturational
order. The functional effects associated with semantic plausibility were investigated by comparing mechanism of cortical thinning in these
responses to sentences about conceptually plausible actions of an animal with those describing con- areas, possibly driven by long-lasting
ceptually implausible actions. Depicted are main effects of syntactic complexity (red), semantic synaptic pruning 96, has been proposed to
implausibility (blue) and the interaction of both factors (purple) (P < 0.01; Bonferroni- and cluster underlie the sophistication of syntactic
size-corrected). In the adult brain, syntax has a distinct spatial representation (left Brodmann area 44 abilities. Future studies combining structural
(BA44)) to that of semantics, and children show slow development towards this segregation. In younger and functional imaging techniques are
children, there is statistical interaction between the pattern of the semantics- and syntax-related needed to corroborate this notion.
haemodynamic responses in the mid and posterior portions of the left superior temporal gyrus (STG).
Ordering phrases and building up
Only 9–10‑year-old children show selectivity for syntax in the left inferior frontal gyrus (IFG); however,
unlike adults, they recruit not only BA44 but also BA45, which suggests that Broca’s area is not yet fully a sentence-level argument structure
functionally specialized at this age9. Adapted with permission from REF. 9, Elsevier. requires a common computational effort
of the left IFG and the left posterior
STG and STS within the dorsal language
semantic processing is segregated from young adulthood. The relationship between network37. A recent study reconstructed
that indicating syntax, as in the adult brain9 phrases is marked by morphological the white-matter fibre tract that directly
(FIG. 2). Replication in follow‑up studies is elements (for example, inflections and connects these areas — the left AF — and
necessary to assess the generalizability of case markers) known as morphosyntactic also the ventral language pathway along the
these data. elements. An ERP study that investigated left IFOF in the four age groups mentioned
the use of morphosyntactic information above10. From this, the individual fractional
Analysis of syntactic relations. The growth during online sentence comprehension anisotropy (FA) values (reflecting structural
of the ability to syntactically categorize showed that although 3‑year-old children connectivity) for each tract of each
heard words enables children in their third can detect grammatical case-marking cues, participant were related to their individual
year of life to analyse syntactic relations it takes until the age of 6 before children sentence comprehension performance.
between immediately neighbouring start to use this information to determine When controlling for the unspecific
phrases. Moreover, unlike newborn babies who is doing what to whom in a sentence93,94. effect of age and the domain-general
and infants, 3–4‑year-old children recruit A combined fMRI–DTI study of 3 groups of influence of verbal working memory, both
not only the left temporal but also the left children aged 3–4, 6–7 or 9–10 years, as well the increase in accuracy and the decrease
inferior frontal cortex when processing as an adult group, compared haemodynamic in response times across development are
sentences9,10,52,92, suggesting that the basic activity underlying the processing of spoken more strongly related to the FA of the AF
functional components for top-down sentences with a canonical word order than to that of the IFOF. Moreover, the
syntax processing are established at this against sentences with a non-canonical correlational relationship between the FA
age. However, the neural and behavioural (but owing to case marking, syntactically of the AF and the behavioural response to
efficiency for computing the syntactic correct) word order. The effect of working syntactically more-complex non-canonical
relationship between remote phrases in memory was removed from the analysis, sentences was markedly stronger than
syntactically complex sentences arises to allow specific investigation of syntactic the relationship between the FA of the AF
gradually, and is only fully established in computation. Children aged 3–7 years and the behavioural response to canonical

328 | MAY 2016 | VOLUME 17 www.nature.com/nrn

©
2
0
1
6
M
a
c
m
i
l
l
a
n
P
u
b
l
i
s
h
e
r
s
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
 PERSPECTIVES

a necessary for extracting basic segmental and

3–4 years 6–7 years 9–10 years Adults
suprasegmental features of speech, which is
carried out in the bilateral superior temporal
and inferior frontal cortices. Through
refining their genetically predisposed
acoustic skills, infants learn to segment the
speech stream into phonological word forms
AF IFOF in their first year of life. Well before their
FA of AF first birthday, infants also start associating
phonological word forms with objects in
FA of IFOF
their environment. By around 12 months,
they can not only access and retrieve
b c items from their lexicon but also begin
40 40
Mean accuracy of responses

Mean accuracy of responses

to semantically categorize these items by
to non-canonical sentences

to non-canonical sentences

recruiting central and posterior parts of the

20 20
left middle and superior temporal cortex.
Around their second birthday, children
0 0
are able to apply their morphosyntactic
knowledge to evaluate the grammatical
–20 –20 3–4 years
compatibility of adjacent phrase categories
6–7 years in the left anterior superior temporal cortex.
–40 –40
9–10 years The top-down processing of semantic
–40 –20 0 20 –40 –20 0 20
Adults and syntactic relations in the left inferior
FA of AF FA of IFOF
frontal cortex emerges in the fourth year of
Figure 3 | The ontogenetic emergence of complex syntax. Processing syntactically complex sen- life but refines only gradually and slowly.
tences requires transfer of syntactic information from the left inferior Nature Reviews
frontal gyrus | Neuroscience
to the left superior Functional selectivity for sentence-level
temporal gyrus, where it can be integrated with semantic information2. In adults, this information semantic information becomes neuro-
transfer depends on a dorsal pathway via the left arcuate fasciulus (AF; blue) but not on a ventral
anatomically separable from functional
pathway via the left inferior fronto-occipital fasciculus (IFOF; green)37. To track the emergence of this
dissociation, the corresponding white-matter fibre tracts were recently reconstructed in four age selectivity for sentence-level syntactical
groups and assessed for their fractional anisotropy (FA) using diffusion-weighted MRI (depicted by information between the ages of 7 and 9, and
wedges; part a). The FA of the AF was more strongly correlated with accuracy in comprehending it is only after the age of 10 that BA44 reaches
syntactically complex, non-canonical sentences (part b) than was the FA of the IFOF (part c). This effect its full specificity and ultimate efficiency
generalized to comparisons between non-canonical and canonical sentences (not shown) and was in processing complex syntax. These
robust to the unspecific effect of age and the domain-general influence of verbal working memory10. maturational trajectories are assumed to be
Adapted from Skeide, M. A., Brauer, J. & Friederici, A. D. Brain functional and structural predictors of shaped by pruning of perisylvian neurons
language performance, Cereb. Cortex, 2015, http://dx.doi.org/10.1093/cercor/bhv042, by permission as well as myelination and growth of their
of Oxford University Press. interconnecting white-matter fibre tracts.

Implications. The present model might have

sentences (FIG. 3). Together, these first results Conclusions implications for practitioners from several
indicate that several other maturational A neurodevelopmental blueprint of fields. It provides an empirical basis for the
mechanisms, including myelination97, axon language. Considering the studies notion that children’s progress in language
growth98 and increasing fibre density 99, outlined above, we present a model of acquisition is coupled to a relatively fixed
are associated with the developmental the ontogeny of the cortical language time course of neurobiological maturation.
refinement of the dorsal syntax network. network. EEG studies on infants and Accordingly, it seems likely that the
The healthy maturation of the AF is a toddlers have provided the main data effectiveness of educational programmes can
necessary precondition for normal language basis for describing the acquisition of be maximized when adapting the complexity
acquisition. Indeed, its disturbance bottom‑up language processing capacities. of the teaching language and educational
might lead to faulty abstract grammatical Owing to methodological challenges, media to the particular neurodevelopmental
representations and might manifest complementary MRI and MEG data are level of the corresponding age group.
itself as specific language impairment 100 currently rare, particularly for the second Our model groups the brain maturation
(see Supplementary information S2 (box)). and third year of life. The exploration of the milestones for normal language acquisition
Taken together, the studies discussed here developmental dynamics of higher-order along a timeline of cognitive development.
suggest that top-down syntax processing sentence-level mechanisms is still in its Thus, it complements current behaviourally
develops by 3 years of age, when sentence beginnings. Nevertheless, these first results informed notions of the developmental
comprehension is still inefficient. Increases allow us to draw a consistent picture that trajectories underlying language acquisition.
in syntax performance continue into can be modified as new data become Speech and language therapists might
adulthood, and are related to the functional available (FIG. 4). consider it to be a useful additional
specialization of the left IFG and the left According to our model, at least 4 weeks source for determining faulty verbal
posterior STG, and to the maturation of before birth, the fetus is already equipped communication skills potentially pointing
their structural connection via the AF. with bottom‑up processing machinery to a language disorder.

NATURE REVIEWS | NEUROSCIENCE VOLUME 17 | MAY 2016 | 329

©
2
0
1
6
M
a
c
m
i
l
l
a
n
P
u
b
l
i
s
h
e
r
s
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
PERSPECTIVES

a Developmental trajectories of language and comprehension

Bottom-up processes
1 Phonological word form detection
2 Morphosyntactic categorization
3 Lexical–semantic categorization

BA44 4 Lexical access and retrieval

BA45 8
7 8 5 Phrase structure reconstruction
pSTG BA44
pSTG 6a 6b 6a Prosodic processing
7 FOP BA41/42 pSTS
BA47 1
1 Top-down processes
5 aSTG 1
6b Prosodic processing
4
aSTS 7 Analysis of semantic relations
3
2
8 Analysis of syntactic relations

b
1
2
3
4
5
6a
6b
7
8
GW 28 0 1 2 3 4 5 6 7 8 9 10
Age (years)

Figure 4 | The evolving cortical circuit underlying language com- Nature

(stages 2 and 6a). At around 32 months old, Reviews | Neuroscience
the reconstruction of phrase
prehension. a | Neural implementation (indicated by circles, shadows, structures elicits adult-like electroencephalography (EEG) responses
and black labelling) of the main processing stages (numbers 1–8) known that are assumed to have sources in left anterior temporal regions 85
from the adult brain. Arrows indicate the assumed flow of information (stage 5). Sentence-level processing of complex syntax develops grad-
along interconnecting white-matter fibre tracts. Bottom‑up processes ually until young adulthood and is driven by the increasing selectivity of
are coloured in green, and top-down processes are coloured in orange. the left inferior frontal cortex and its connection to the posterior
b | A model of the language acquisition timeline (age in years). The superior temporal cortex via the arcuate fasciculus9,10 (stage 8). From
earliest manifestations (starting points of the lines) and first appearance 6 months onwards, children access and retrieve lexicon entries; that is,
of adult-like functions (end points of the lines) of different neural pro- memorized associations between phonological word forms and objects
cessing milestones, at the group level, are shown. Note that these par- in their environment. Around their first birthday, they sort lexical items
ticular processes do not emerge and disappear abruptly, but do so into conceptual categories by recruiting central and posterior parts of
grad­ually. Moreover, they are subject to inter-individual variation. From the left middle and superior temporal cortex11,72 (stage 3). By 2 years of
gestational week 28 (GW 28) onwards, at the latest, infants are able to age, children can already evaluate the semantic compatibility of ele-
extract basic segmental (phonological) and suprasegmental (prosodic) ments in a sentence74 (stage 4), and this semantic information serves as
features of speech by recruiting the bilateral superior temporal sulcus a cue for understanding syntactically complex sentences until around
(STS)12. In the first 9 months of life, they have to rely on prosodic cues (for 9 years of age9 (stage 7). Prosodic processing at the sentence level is
example, accentuation) to detect phonological word forms, but this documented in 3‑year-old children, but it is only at around 6 years of age
dependency is no longer present by around 12 months of age63 (stage 1). that children can recognize phrases in sentences without relying on
Prosodic cues also enable 6‑month-old infants to categorize speech pause information81 (stage 6b). aSTG, anterior superior temporal gyrus;
inputs according to morphosyntactic criteria; for example, into noun aSTS, anterior superior temporal sulcus; BA, Brodmann area; FOP, frontal
and verb phrases78. However, by around the second birthday, prosodic opercular cortex; pSTG, posterior superior temporal gyrus; pSTS, posterior
information is no longer used for morphosyntactic categorization84 superior temporal sulcus.

To date, neuropaediatricians have to Outstanding questions. Several important However, the role of these structures in
rely largely on adult data when planning research questions remain to be the acquisition of the first language is not
surgical operations in children. The current answered in follow‑up studies on the yet understood.
model could foster efforts to design better functional specialization and structural Furthermore, considering the different
methods for localizing specific language maturation of the language system. linguistic levels of language processing,
functions in the individual developing For example, subcortical structures, substantial progress has been made in
brain. Novel diagnostic tools might make in particular the left caudate nucleus uncovering the neural development of
it possible to more accurately assess and the thalamus, are known to support phonology, syntax and semantics, but hardly
the risk of language impairment that the acquisition of a second language, anything is known about the neural basis of
follows a surgical procedure in young presumably by regulating adaptation emerging pragmatic skills — that is, using
neurological patients. to increasing processing demands101,102. contextual information for inferring the

330 | MAY 2016 | VOLUME 17 www.nature.com/nrn

©
2
0
1
6
M
a
c
m
i
l
l
a
n
P
u
b
l
i
s
h
e
r
s
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
 PERSPECTIVES

meaning of utterances. This field could be temporal structure revealed by responses to sound connections revealed by combined fMRI/DTI. Cereb.
quilts. Nat. Neurosci. 18, 903–911 (2015). Cortex 22, 191–200 (2012).
explored by combining brain measures with 17. Herrmann, B., Maess, B., Hahne, A., Schroger, E. 43. Schnupp, J., Nelken, I. & King, A. Auditory
behavioural paradigms that investigate how & Friederici, A. D. Syntactic and auditory spatial Neuroscience: Making Sense of Sound (MIT Press,
processing in the human temporal cortex: an MEG 2011).
children disambiguate sentences such as study. Neuroimage 57, 624–633 (2011). 44. Gomez, D. M. et al. Language universals at birth.
“The window is open”, which can be a simple 18. MacGregor, L. J., Pulvermuller, F., van Casteren, M. Proc. Natl Acad. Sci. USA 111, 5837–5841 (2014).
& Shtyrov, Y. Ultra-rapid access to words in the brain. 45. Partanen, E. et al. Learning-induced neural plasticity
statement, an expression of being cold or a Nat. Commun. 3, 711 (2012). of speech processing before birth. Proc. Natl Acad.
request to close the window, depending on 19. Visser, M., Jefferies, E. & Lambon Ralph, M. A. Sci. USA 110, 15145–15150 (2013).
Semantic processing in the anterior temporal lobes: 46. Pena, M., Werker, J. F. & Dehaene-Lambertz, G.
the communicational situation. a meta-analysis of the functional neuroimaging Earlier speech exposure does not accelerate speech
Finally, extending the perspective from literature. J. Cogn. Neurosci. 22, 1083–1094 (2010). acquisition. J. Neurosci. 32, 11159–11163 (2012).
20. Mitchell, T. M. et al. Predicting human brain activity 47. Pena, M., Pittaluga, E. & Mehler, J. Language
the neurobiology to the biology of language, associated with the meanings of nouns. Science 320, acquisition in premature and full-term infants.
there is now a consensus that both nature 1191–1195 (2008). Proc. Natl Acad. Sci. USA 107, 3823–3828 (2010).
21. Binder, J. R., Desai, R. H., Graves, W. W. & Conant, L. L. 48. Benasich, A. A., Choudhury, N. A., Realpe-Bonilla, T.
and nurture make us proficient language Where is the semantic system? A critical review and & Roesler, C. P. Plasticity in developing brain: active
users. But whereas the environmental factors meta-analysis of 120 functional neuroimaging studies. auditory exposure impacts prelinguistic acoustic
Cereb. Cortex 19, 2767–2796 (2009). mapping. J. Neurosci. 34, 13349–13363 (2014).
that shape the language-proficient brain 22. Brennan, J. et al. Syntactic structure building in the 49. Werker, J. F. & Tees, R. C. Cross-language speech-
have been studied extensively in the past few anterior temporal lobe during natural story listening. perception: evidence for perceptual reorganization
Brain Lang. 120, 163–173 (2012). during the 1st year of life. Infant Behav. Dev. 7, 49–63
years103, the genetic contributions to specific 23. Snijders, T. M. et al. Retrieval and unification of (1984).
language skills remain largely unknown. syntactic structure in sentence comprehension: 50. Kuhl, P. K. Early language acquisition: cracking the
an fMRI study using word-category ambiguity. speech code. Nat. Rev. Neurosci. 5, 831–843 (2004).
Future work needs to specify systematically Cereb. Cortex 19, 1493–1503 (2009). 51. Werker, J. F. & Hensch, T. K. Critical periods in speech
how genetic variation and environmental 24. Newman, S. D., Ikuta, T. & Burns, T. Jr. The effect of perception: new directions. Annu. Rev. Psychol. 66,
semantic relatedness on syntactic analysis: An fMRI 173–196 (2015).
influences interact to give rise to the study. Brain Lang. 113, 51–58 (2010). 52. Dehaene-Lambertz, G. et al. Functional organization
language-ready brain. 25. Obleser, J., Wise, R. J., Dresner, M. A. & Scott, S. K. of perisylvian activation during presentation of
Functional integration across brain regions improves sentences in preverbal infants. Proc. Natl Acad. Sci.
speech perception under adverse listening conditions. USA 103, 14240–14245 (2006).
Michael A. Skeide and Angela D. Friederici are at the
J. Neurosci. 27, 2283–2289 (2007). 53. Dubois, J. et al. Exploring the early organization and
Department of Neuropsychology, Max Planck Institute 26. Rodd, J. M., Davis, M. H. & Johnsrude, I. S. The neural maturation of linguistic pathways in the human infant
for Human Cognitive and Brain Sciences, mechanisms of speech comprehension: fMRI studies of brain. Cereb. Cortex http://dx.doi.org/10.1093/cercor/
Stephanstrasse 1a, 04103 Leipzig, Germany. semantic ambiguity. Cereb. Cortex 15, 1261–1269 bhv082 (2015).
(2005). 54. Bruderer, A. G., Danielson, D. K., Kandhadai, P. &
[email protected];
27. Weiller, C., Bormann, T., Saur, D., Musso, M. & Werker, J. F. Sensorimotor influences on speech
[email protected] Rijntjes, M. How the ventral pathway got lost: and perception in infancy. Proc. Natl Acad. Sci. USA 112,
what its recovery might mean. Brain Lang. 118, 13531–13536 (2015).
doi:10.1038/nrn.2016.23 29–39 (2011). 55. Mottonen, R. & Watkins, K. E. Motor representations
Published online 4 Apr 2016 28. Lau, E. F., Phillips, C. & Poeppel, D. A cortical network of articulators contribute to categorical perception
for semantics: (de)constructing the N400. Nat. Rev. of speech sounds. J. Neurosci. 29, 9819–9825
1. Berwick, R. C., Friederici, A. D., Chomsky, N. & Neurosci. 9, 920–933 (2008). (2009).
Bolhuis, J. J. Evolution, brain, and the nature of 29. Friederici, A. D. & Gierhan, S. M. The language 56. Friederici, A. D., Mueller, J. L. & Oberecker, R.
language. Trends Cogn. Sci. 17, 89–98 (2013). network. Curr. Opin. Neurobiol. 23, 250–254 (2013). Precursors to natural grammar learning: preliminary
2. Friederici, A. D. The cortical language circuit: from 30. Makuuchi, M., Bahlmann, J., Anwander, A. & evidence from 4‑month-old infants. PLoS ONE 6,
auditory perception to sentence comprehension. Friederici, A. D. Segregating the core computational e17920 (2011).
Trends Cogn. Sci. 16, 262–268 (2012). faculty of human language from working memory. 57. Gervain, J., Berent, I. & Werker, J. F. Binding at birth:
3. Hagoort, P. MUC (Memory, Unification, Control) and Proc. Natl Acad. Sci. USA 106, 8362–8367 (2009). the newborn brain detects identity relations and
beyond. Front. Psychol. 4, 416 (2013). 31. Meyer, L., Obleser, J., Anwander, A. & Friederici, A. D. sequential position in speech. J. Cogn. Neurosci. 24,
4. Bornkessel-Schlesewsky, I. & Schlesewsky, M. Linking ordering in Broca’s area to storage in left 564–574 (2012).
Reconciling time, space and function: a new dorsal– temporo-parietal regions: the case of sentence 58. Friederici, A. D. Language development and the
ventral stream model of sentence comprehension. processing. Neuroimage 62, 1987–1998 (2012). ontogeny of the dorsal pathway. Front. Evol. Neurosci.
Brain Lang. 125, 60–76 (2013). 32. Friederici, A. D. The brain basis of language 4, 3 (2012).
5. Friederici, A. D. Neurophysiological markers of early processing: from structure to function. Physiol. Rev. 59. Gervain, J., Macagno, F., Cogoi, S., Pena, M. &
language acquisition: from syllables to sentences. 91, 1357–1392 (2011). Mehler, J. The neonate brain detects speech structure.
Trends Cogn. Sci. 9, 481–488 (2005). 33. Dapretto, M. & Bookheimer, S. Y. Form and content: Proc. Natl Acad. Sci. USA 105, 14222–14227 (2008).
6. Friederici, A. D. The neural basis of language dissociating syntax and semantics in sentence 60. Homae, F., Watanabe, H., Nakano, T., Asakawa, K.
development and its impairment. Neuron 52, comprehension. Neuron 24, 427–432 (1999). & Taga, G. The right hemisphere of sleeping infant
941–952 (2006). 34. Dick, A. S. & Tremblay, P. Beyond the arcuate perceives sentential prosody. Neurosci. Res. 54,
7. Kuhl, P. K. Brain mechanisms in early language fasciculus: consensus and controversy in the 276–280 (2006).
acquisition. Neuron 67, 713–727 (2010). connectional anatomy of language. Brain 135, 61. Seltzer, L. J., Ziegler, T. E. & Pollak, S. D. Social
8. Perani, D. et al. Neural language networks at birth. 3529–3550 (2012). vocalizations can release oxytocin in humans.
Proc. Natl Acad. Sci. USA 108, 16056–16061 (2011). 35. den Ouden, D. B. et al. Network modulation Proc. Biol. Sci. 277, 2661–2666 (2010).
9. Skeide, M. A., Brauer, J. & Friederici, A. D. Syntax during complex syntactic processing. Neuroimage 59, 62. Friederici, A. D., Friedrich, M. & Christophe, A. Brain
gradually segregates from semantics in the developing 815–823 (2012). responses in 4‑month-old infants are already language
brain. Neuroimage 100, 106–111 (2014). 36. Makuuchi, M. & Friederici, A. D. Hierarchical specific. Curr. Biol. 17, 1208–1211 (2007).
10. Skeide, M. A., Brauer, J. & Friederici, A. D. Brain functional connectivity between the core language 63. Mannel, C. & Friederici, A. D. Accentuate or repeat?
functional and structural predictors of language system and the working memory system. Cortex 49, Brain signatures of developmental periods in infant
performance. Cereb. Cortex http://dx.doi.org/10.1093/ 2416–2423 (2013). word recognition. Cortex 49, 2788–2798 (2013).
cercor/bhv042 (2015). 37. Wilson, S. M. et al. Syntactic processing depends on 64. Bosseler, A. N. et al. Theta brain rhythms index
11. Travis, K. E. et al. Spatiotemporal neural dynamics of dorsal language tracts. Neuron 72, 397–403 (2011). perceptual narrowing in infant speech perception.
word understanding in 12- to 18‑month-old-infants. 38. Saur, D. et al. Ventral and dorsal pathways Front. Psychol. 4, 690 (2013).
Cereb. Cortex 21, 1832–1839 (2011). for language. Proc. Natl Acad. Sci. USA 105, 65. Ortiz-Mantilla, S., Hamalainen, J. A., Musacchia, G.
12. Mahmoudzadeh, M. et al. Syllabic discrimination in 18035–18040 (2008). & Benasich, A. A. Enhancement of gamma oscillations
premature human infants prior to complete formation 39. Weiller, C., Musso, M., Rijntjes, M. & Saur, D. indicates preferential processing of native over foreign
of cortical layers. Proc. Natl Acad. Sci. USA 110, Please don’t underestimate the ventral pathway in phonemic contrasts in infants. J. Neurosci. 33,
4846–4851 (2013). language. Trends Cogn. Sci. 13, 369–370 (2009). 18746–18754 (2013).
13. Fedorenko, E. & Thompson-Schill, S. L. Reworking the 40. Friederici, A. D., Bahlmann, J., Heim, S., 66. Kuhl, P. K., Ramirez, R. R., Bosseler, A., Lin, J. F. &
language network. Trends Cogn. Sci. 18, 120–126 Schubotz, R. I. & Anwander, A. The brain differentiates Imada, T. Infants’ brain responses to speech suggest
(2014). human and non-human grammars: functional analysis by synthesis. Proc. Natl Acad. Sci. USA 111,
14. DeWitt, I. & Rauschecker, J. P. Phoneme and word localization and structural connectivity. Proc. Natl 11238–11245 (2014).
recognition in the auditory ventral stream. Proc. Natl Acad. Sci. USA 103, 2458–2463 (2006). 67. Gervain, J. Plasticity in early language acquisition: the
Acad. Sci. USA 109, E505–E514 (2012). 41. Fruhholz, S., Ceravolo, L. & Grandjean, D. Specific effects of prenatal and early childhood experience.
15. Hickok, G. & Poeppel, D. The cortical organization of brain networks during explicit and implicit decoding Curr. Opin. Neurobiol. 35, 13–20 (2015).
speech processing. Nat. Rev. Neurosci. 8, 393–402 of emotional prosody. Cereb. Cortex 22, 1107–1117 68. Pons, F., Lewkowicz, D. J., Soto-Faraco, S. &
(2007). (2012). Sebastian-Galles, N. Narrowing of intersensory speech
16. Overath, T., McDermott, J. H., Zarate, J. M. & 42. Ethofer, T. et al. Emotional voice areas: anatomic perception in infancy. Proc. Natl Acad. Sci. USA 106,
Poeppel, D. The cortical analysis of speech-specific location, functional properties, and structural 10598–10602 (2009).

NATURE REVIEWS | NEUROSCIENCE VOLUME 17 | MAY 2016 | 331

©
2
0
1
6
M
a
c
m
i
l
l
a
n
P
u
b
l
i
s
h
e
r
s
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.
PERSPECTIVES

69. Kuhl, P. K. et al. Infants show a facilitation effect for 84. Oberecker, R. & Friederici, A. D. Syntactic event- 97. Gupta, N. et al. Neural stem cell engraftment and
native language phonetic perception between 6 and related potential components in 24‑month-olds’ myelination in the human brain. Sci. Transl Med. 4,
12 months. Dev. Sci. 9, F13–F21 (2006). sentence comprehension. Neuroreport 17, 1017–1021 155ra137 (2012).
70. Wartenburger, I. et al. The processing of prosody: (2006). 98. Paus, T. Growth of white matter in the adolescent
Evidence of interhemispheric specialization at the age 85. Oberecker, R., Friedrich, M. & Friederici, A. D. Neural brain: myelin or axon? Brain Cogn. 72, 26–35 (2010).
of four. Neuroimage 34, 416–425 (2007). correlates of syntactic processing in two-year-olds. 99. Scholz, J., Klein, M. C., Behrens, T. E. &
71. Bergelson, E. & Swingley, D. At 6–9 months, human J. Cogn. Neurosci. 17, 1667–1678 (2005). Johansen‑Berg, H. Training induces changes
infants know the meanings of many common nouns. 86. Tyler, L. K. et al. Left inferior frontal cortex and syntax: in white-matter architecture. Nat. Neurosci. 12,
Proc. Natl Acad. Sci. USA 109, 3253–3258 (2012). function, structure and behaviour in patients with left 1370–1371 (2009).
72. Friedrich, M., Wilhelm, I., Born, J. & Friederici, A. D. hemisphere damage. Brain 134, 415–431 (2011). 100. van der Lely, H. K. & Pinker, S. The biological basis of
Generalization of word meanings during infant sleep. 87. Friedrich, M. & Friederici, A. D. Lexical priming and language: insight from developmental grammatical
Nat. Commun. 6, 6004 (2015). semantic integration reflected in the event-related impairments. Trends Cogn. Sci. 18, 586–595 (2014).
73. Fiebach, C. J., Friederici, A. D., Muller, K., von potential of 14‑month-olds. Neuroreport 16, 653–656 101. Wartenburger, I. et al. Early setting of grammatical
Cramon, D. Y. & Hernandez, A. E. Distinct brain (2005). processing in the bilingual brain. Neuron 37, 159–170
representations for early and late learned words. 88. Friedrich, M. & Friederici, A. D. Phonotactic (2003).
Neuroimage 19, 1627–1637 (2003). knowledge and lexical-semantic processing in one- 102. Crinion, J. et al. Language control in the bilingual
74. Friedrich, M. & Friederici, A. D. Maturing brain year-olds: brain responses to words and nonsense brain. Science 312, 1537–1540 (2006).
mechanisms and developing behavioral language words in picture contexts. J. Cogn. Neurosci. 17, 103. Tomasello, M. Constructing a Language: a Usage-
skills. Brain Lang. 114, 66–71 (2010). 1785–1802 (2005). Based Theory of Language Acquisition (Harvard Univ.
75. Hohle, B., Weissenborn, E., Kiefer, D., Schulz, A. & 89. Friedrich, M. & Friederici, A. D. Semantic sentence Press, 2003).
Schmitz, M. Functional elements in infants’ speech processing reflected in the event-related potentials 104. Broca, P. Localisations des fonctions cérébrales. Siège
processing: The role of determiners in the syntactic of one- and two-year-old children. Neuroreport 16, de la faculté du langage articulé. Bull. Société d’
categorization of lexical elements. Infancy 5, 341–353 1801–1804 (2005). Anthropol. Paris 4, 200–204 (in French) (1863).
(2004). 90. Nunez, S. C. et al. fMRI of syntactic processing in 105. Hagoort, P. & Levelt, W. J. Neuroscience. The speaking
76. Shi, R. S. & Melancon, A. Syntactic categorization in typically developing children: structural correlates brain. Science 326, 372–373 (2009).
French-learning infants. Infancy 15, 517–533 (2010). in the inferior frontal gyrus. Dev. Cogn. Neurosci. 1, 106. Sahin, N. T., Pinker, S., Cash, S. S., Schomer, D.
77. Dautriche, I., Swingley, D. & Christophe, A. Learning 313–323 (2011). & Halgren, E. Sequential processing of lexical,
novel phonological neighbors: syntactic category 91. Brauer, J. & Friederici, A. D. Functional neural grammatical, and phonological information within
matters. Cognition 143, 77–86 (2015). networks of semantic and syntactic processes in the Broca’s area. Science 326, 445–449 (2009).
78. Soderstrom, M., Seidl, A., Nelson, D. G. K. & developing brain. J. Cogn. Neurosci. 19, 1609–1623 107. Flinker, A. et al. Redefining the role of Broca’s area in
Jusczyk, P. W. The prosodic bootstrapping of phrases: (2007). speech. Proc. Natl Acad. Sci. USA 112, 2871–2875
evidence from prelinguistic infants. J. Mem. Lang. 49, 92. Dehaene-Lambertz, G., Dehaene, S. & Hertz- (2015).
249–267 (2003). Pannier, L. Functional neuroimaging of speech 108. Dehaene, S., Cohen, L., Morais, J. & Kolinsky, R.
79. Mannel, C. & Friederici, A. D. Pauses and intonational perception in infants. Science 298, 2013–2015 Illiterate to literate: behavioural and cerebral changes
phrasing: ERP studies in 5‑month-old German infants (2002). induced by reading acquisition. Nat. Rev. Neurosci.
and adults. J. Cogn. Neurosci. 21, 1988–2006 (2009). 93. Schipke, C. S., Friederici, A. D. & Oberecker, R. Brain 16, 234–244 (2015).
80. Mannel, C. & Friederici, A. D. Intonational phrase responses to case-marking violations in German
structure processing at different stages of syntax preschool children. Neuroreport 22, 850–854 (2011). Acknowledgements
acquisition: ERP studies in 2-, 3-, and 6‑year-old 94. Schipke, C. S., Knoll, L. J., Friederici, A. D. & This work was supported by a grant from the European
children. Dev. Sci. 14, 786–798 (2011). Oberecker, R. Preschool children’s interpretation of Research Council (ERC‑2010‑AdG 20100407, NEUROSYNTAX)
81. Mannel, C., Schipke, C. S. & Friederici, A. D. The role object-initial sentences: neural correlates of their awarded to A.D.F.
of pause as a prosodic boundary marker: language behavioral performance. Dev. Sci. 15, 762–774
ERP studies in German 3- and 6‑year-olds. Dev. Cogn. (2012). Competing interests statement
Neurosci. 5, 86–94 (2013). 95. Fengler, A., Meyer, L. & Friederici, A. D. Brain The authors declare no competing interests.
82. Sammler, D., Kotz, S. A., Eckstein, K., Ott, D. V. M. & structural correlates of complex sentence
Friederici, A. D. Prosody meets syntax: the role of the comprehension in children. Dev. Cogn. Neurosci. 15,
corpus callosum. Brain 133, 2643–2655 (2010). 48–57 (2015). SUPPLEMENTARY INFORMATION
83. van Heugten, M. & Christophe, A. Infants’ acquisition 96. Petanjek, Z. et al. Extraordinary neoteny of synaptic See online article: S1 (box) | S2 (box)
of grammatical gender dependencies. Infancy 20, spines in the human prefrontal cortex. Proc. Natl ALL LINKS ARE ACTIVE IN THE ONLINE PDF
675–683 (2015). Acad. Sci. USA 108, 13281–13286 (2011).

332 | MAY 2016 | VOLUME 17 www.nature.com/nrn

©
2
0
1
6
M
a
c
m
i
l
l
a
n
P
u
b
l
i
s
h
e
r
s
L
i
m
i
t
e
d
.
A
l
l
r
i
g
h
t
s
r
e
s
e
r
v
e
d
.