NE37CH18-Hagoort ARI 30 June 2014 9:10

ANNUAL
REVIEWS Further The Neurobiology of Language
Downloaded from www.annualreviews.org. ONOS (39152) IP: 202.41.10.108 On: Tue, 22 Jul 2025 11:51:37

Click here for quick links to

Annual Reviews content online,
including:
Beyond Single Words
• Other articles in this volume
• Top cited articles
• Top downloaded articles
Peter Hagoort1,2 and Peter Indefrey2,3
• Our comprehensive search 1
Max Planck Institute for Psycholinguistics, 6525 XD Nijmegen, The Netherlands;
email: peter.hagoort@donders.ru.nl
2
Donders Institute for Brain, Cognition and Behaviour, Radboud University Nijmegen,
6525 EN Nijmegen, The Netherlands
3
Heinrich Heine University, 40225 Düsseldorf, Germany; email: indefrey@phil.hhu.de

Annu. Rev. Neurosci. 2014. 37:347–62 Keywords

The Annual Review of Neuroscience is online at syntactic processing, semantic processing, neuropragmatics, Broca’s area,
neuro.annualreviews.org
Wernicke’s area, meta-analysis
This article’s doi:
10.1146/annurev-neuro-071013-013847 Abstract
Copyright  c 2014 by Annual Reviews. A hallmark of human language is that we combine lexical building blocks
All rights reserved
retrieved from memory in endless new ways. This combinatorial aspect of
language is referred to as unification. Here we focus on the neurobiological
First published online as a Review in Advance on infrastructure for syntactic and semantic unification. Unification is char-
June 2, 2014
acterized by a high-speed temporal profile including both prediction and
Erratum
integration of retrieved lexical elements. A meta-analysis of numerous neu-
roimaging studies reveals a clear dorsal/ventral gradient in both left inferior
frontal cortex and left posterior temporal cortex, with dorsal foci for syn-
tactic processing and ventral foci for semantic processing. In addition to
core areas for unification, further networks need to be recruited to realize
language-driven communication to its full extent. One example is the theory
of mind network, which allows listeners and readers to infer the intended
message (speaker meaning) from the coded meaning of the linguistic ut-
terance. This indicates that sensorimotor simulation cannot handle all of
language processing.

347
NE37CH18-Hagoort ARI 30 June 2014 9:10

Contents
INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
Unification and Prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 348
Shared Circuitry for Comprehension and Production . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 350
SYNTACTIC VERSUS SEMANTIC UNIFICATION: A META-ANALYSIS . . . . . . 351
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 352
Summary and Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 356
Downloaded from www.annualreviews.org. ONOS (39152) IP: 202.41.10.108 On: Tue, 22 Jul 2025 11:51:37

BEYOND CORE AREAS FOR LANGUAGE: THE CASE

OF NEUROPRAGMATICS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 358
CONCLUSION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 359

INTRODUCTION
Language processing is more than memory retrieval and more than the simple concatenation of
retrieved lexical items. The expressive power of human language is based on the ability to combine
elements from memory in novel ways. This notion led the linguist Wilhelm von Humboldt (1829)
to his famous claim that language “makes infinite use of finite means.” This central feature of
the human language system is, however, not represented in the classical Wernicke-Lichtheim-
Geschwind (WLG) model of the neural architecture of language. This model did not go beyond
the single word level (Hagoort 2013, Levelt 2013). Even today a large proportion of neuroimaging
research on language focuses on processing at or below this level. For example, one of the most in-
fluential recent models of the neurobiology of language, that of Hickok & Poeppel (2007), focuses
mainly on speech processing at lexical and sublexical levels. Our focus here is instead on processes
beyond the retrieval of lexical information. Clearly, retrieval of the lexical building blocks for
language (e.g., phonological, morphological, syntactic building blocks) is a crucial component of
human language processing (Hagoort 2005, 2013). But equally important is the process of deriving
new and complex meaning from the lexical building blocks. This process is referred to in different
ways. Here we use the term unification to indicate the real-time assembly of the pieces retrieved
from memory into larger structures, with contributions from context. Classic psycholinguistic
studies of unification had a strong focus on syntactic analysis (parsing). Unification operations
take place at the syntactic processing level, but not only there (cf. Jackendoff 2002, 2007). At the
semantic and phonological levels, too, conceptual and lexical elements are combined and inte-
grated into larger structures. It is therefore useful to distinguish between syntactic, semantic, and
phonological unification (Hagoort 2005, 2013). Before discussing the neurobiological infrastruc-
ture for unification in more detail, we discuss the temporal dynamics of the operations supported
by this infrastructure.

Unification and Prediction

One of the most remarkable characteristics of speaking and listening is the speed at which it occurs.
Speakers produce easily between 2 and 5 words per second, information that must be decoded by
the listener within roughly the same time frame. Given the rate of typical speech we can deduce
that word recognition is extremely fast and efficient, taking no more than 200–300 ms. Language
processing is highly incremental, with a continuous interaction between the preceding context
and the ongoing lexical retrieval operations (Zwitserlood 1989). Here prediction and postlexical

348 Hagoort · Indefrey

NE37CH18-Hagoort ARI 30 June 2014 9:10

integration might work together (cf. Kutas et al. 2015). Electrophysiological data from event-
related brain potential (ERP) studies on language provide relevant insights about the time course
of the incremental updating of the input representation.
Considering that the acoustic duration of many words is on the order of a few hundred mil-
liseconds, the immediacy of the electrophysiological language-related effects is remarkable. For
instance, the early left anterior negativity (ELAN), a syntax-related effect (Friederici et al. 2003),
starts at ∼100–150 ms after onset of the acoustic word (but see Steinhauer & Drury 2012 for a
critical discussion of the ELAN). The onset of the N400 is at ∼250 ms, and another language-
relevant ERP, the so-called P600, usually starts at ∼500 ms. Thus some of these effects occur well
Downloaded from www.annualreviews.org. ONOS (39152) IP: 202.41.10.108 On: Tue, 22 Jul 2025 11:51:37

before the end of a spoken word. Classifying visual input (e.g., a picture) as depicting an animate or
inanimate entity takes the brain ∼150 ms (Thorpe et al. 1996). Roughly the same amount of time is
needed to classify orthographic input as a letter (Grainger et al. 2008). If we take this time estimate
as our reference point, the early appearance of electrophysiological responses to a spoken word is
a salient feature of word processing. In physiological terms, the earliness of some of these brain
responses suggests that they are not influenced by long-range recurrent feedback to the primary
and secondary auditory cortices involved in first-pass acoustic and phonological analysis. Recent
modeling work suggests that early ERP effects are best explained by a model with feedforward
connections only. Backward connections become essential only after 220 ms (Garrido et al. 2007).
The effects of backward connections are, therefore, not manifest in the latency range of early ERP
effects such as the ELAN because not enough time has passed for return activity from higher levels
to auditory cortex. In the case of speech, the N400 follows the word recognition points closely in
time, which suggests that what occurs on-line (that is, in real time) in language comprehension is
presumably based partly on predictive processing. Under many circumstances, there is simply not
enough time for top-down feedback to exert control over a preceding bottom-up analysis. Very
likely, lexical, semantic, and syntactic cues conspire to predict sometimes very detailed charac-
teristics of the next anticipated word, including its syntactic and semantic makeup. A mismatch
between contextual prediction and the output of bottom-up analysis results in an immediate brain
response recruiting additional processing resources to salvage the on-line interpretation process.
Recent ERP studies have provided evidence that context can indeed result in predictions about
the syntactic features (i.e., grammatical gender; Van Berkum et al. 2005) and word form (DeLong
et al. 2005) of the next word. Lau et al. (2006) showed that the ELAN elicited by a word category
violation was modulated by the strength of the expectation for a particular word category in the
relevant syntactic slot. More general event knowledge or semantic domains seem to assist with
prediction of upcoming information (see Metusalem et al. 2012, Szewczyk & Schriefers 2013,
Kutas et al. 2015 for a review of the relevant literature). Whereas the evidence for prediction
is strong in the case of language comprehension, the role of prediction in language production
(speaking) is less clear. The listener can, at best, guess (predict) what the speaker will say next, but
the speaker herself knows where her speech is headed. Hence, in this case, planning an utterance
seems to coincide with prediction (Meyer & Hagoort 2013).
Unification refers to the ongoing combinatorial operations in language, which include both
context-driven predictions and integration of the lexical information into a representation that
spans the whole utterance. The difference is that predictions work forward in time, whereas
integration works backward in time; in the latter case, information is integrated into a context that
preceded the currently processed lexical item.
Unification is not restricted to linguistic building blocks. ERP studies have shown that non-
linguistic information also contributes to unification. Here one can think of information about
the speaker (Van Berkum et al. 2008), co-speech gestures (Willems et al. 2007), or world knowl-
edge (Hagoort et al. 2004). These studies found the same effects for nonlinguistic compared with

www.annualreviews.org • Language Beyond Single Words 349

NE37CH18-Hagoort ARI 30 June 2014 9:10

linguistic information on the amplitude and latency of ERPs such as the N400. These data suggest
that both linguistic and nonlinguistic contexts have an immediate impact on the interpretation of
an utterance (Hagoort & van Berkum 2007).

Shared Circuitry for Comprehension and Production

The WLG model strictly divided between areas involved in language production (Broca’s area)
and language comprehension (Wernicke’s area). It has become clear that this division is incorrect.
Downloaded from www.annualreviews.org. ONOS (39152) IP: 202.41.10.108 On: Tue, 22 Jul 2025 11:51:37

Lesions in Broca’s area impair not only language production but also language comprehension
(Caramazza & Zurif 1976), whereas lesions in Wernicke’s area also affect language production.
More recent neuroimaging studies provided further evidence that the classical view on the role
of these regions is no longer tenable. Central aspects of language production and comprehension
are subserved by shared neural circuitry (Menenti et al. 2011, Segaert et al. 2012). For instance,
Segaert et al. (2012) investigated whether the neurobiological substrate for coding and processing
syntactic representations is shared between speaking and listening. To ensure that not just the
same brain regions but in fact the same neuronal populations within these regions are involved
in both modalities, these investigators used a repetition suppression design (Grill-Spector et al.
2006). For repeated syntactic structures (e.g., actives, passives) within and between modalities
(i.e., speaking and listening), they found the same suppression effects for within- and for between-
modality repetitions. These effects were localized in the left middle temporal gyrus (MTG),
the left inferior frontal gyrus (IFG), and the left supplementary motor area (see Supplemental
Supplemental Material Figure 1. Follow the Supplemental Material link from the Annual Reviews home page at
http://www.annualreviews.org).
Given that speaking and listening have different processing requirements, the same areas may
not be recruited to the same degree in these two cases (Indefrey et al. 2004). This difference
between speaking and listening, however, does not detract from the finding that when needed, the
same areas subserve unification operations in language production and language comprehension.
Even less clear is the degree of separation between networks for syntactic and semantic uni-
fication. Separation of these networks is nontrivial because syntactic differences are not without
semantic consequences. For instance, sentences containing an object-relative clause (e.g., “The
reporter who the senator attacked admitted the error”) are found to be more difficult to under-
stand than sentences with a subject-relative clause (e.g., “The reporter who attacked the senator
admitted the error”). This increased difficulty is usually attributed to the difference in the syntac-
tic structure of object and subject relative clauses. Chen and coworkers (2006), however, showed
that increases in activation of Broca’s area for object compared with subject relative clauses are
observed only when the subject of the relative clause is inanimate and the subject of the main
clause is animate (e.g., “The golfer that the lightning struck”) and not for the syntactically identi-
cal sentences with animate relative clause subjects (e.g., “The wood that the man chopped heated
the cabin”). Chen et al. argue that this finding rules out hypotheses based on syntactic struc-
ture complexity and rather suggests that activation of Broca’s area reflects the relative difficulty
of thematic role assignment; inanimate referents were dispreferred agents of an action and ani-
mate referents dispreferred undergoers of action (see also Kuperberg et al. 2008). A difference in
syntactic structure clearly has immediate consequences for mapping grammatical roles (subject,
object) onto thematic roles (agent, undergoer) and hence for the event structure depicted in these
sentences. The event structure is part of the conceptual/semantic representation of the sentence.
Despite the intertwined nature of syntactic and semantic unification, for this review we attempt to
segregate these two types of unification as much as possible using data from a new meta-analysis
of numerous functional imaging studies.

350 Hagoort · Indefrey

NE37CH18-Hagoort ARI 30 June 2014 9:10

SYNTACTIC VERSUS SEMANTIC UNIFICATION: A META-ANALYSIS

So far, the available information is most clear for the areas involved in syntactic analysis. In a recent
meta-analysis of more than 80 neuroimaging studies on sentence processing, Indefrey (2012) iden-
tified consistent activations in a comparison of sentences that were syntactically more demanding
and control sentences that were syntactically less demanding (e.g., object-relative versus subject-
relative sentences). Two regions were systematically more strongly activated when participants
were reading or listening to syntactically demanding sentences. The one region is Broca’s area in
the left inferior frontal gyrus (LIFG), more specifically the pars opercularis [Brodmann area (BA)
Downloaded from www.annualreviews.org. ONOS (39152) IP: 202.41.10.108 On: Tue, 22 Jul 2025 11:51:37

44] and the pars triangularis (BA 45). The other region includes the posterior parts of the left
superior and middle temporal gyri. These results are remarkably consistent with those from the
syntactic adaptation study by Segaert et al. (2012), with the outcome of a representational similar-
ity analysis of magnetoencephalography (MEG) data (Tyler et al. 2013), and with a recent analysis
of lesion data in patients with a syntactic comprehension deficit (Tyler et al. 2011). Although the
specific contributions of these areas to syntactic analysis is a matter of debate, the involvement of
the LIFG and left posterior MTG cannot be denied.
Hereafter we present a new meta-analysis of neuroimaging studies on sentence processing
to establish, among other things, the degree to which the network for semantic unification
can be segregated from the network for syntactic unification. We analyzed 151 hemodynamic
studies on sentence processing that reported activated brain regions with Talairach coordinates
or Montreal Neurological Institute (MNI) coordinates (see Supplemental Table 1). A subset Supplemental Material
of 85 studies contrasting sentence-level processing with lower-level control conditions, or syn-
tactically more demanding sentences with less demanding sentences, were taken from previous
meta-analyses, such as the one discussed above (Indefrey 2011, 2012). To identify additional
studies on sentence-level semantic processing, we conducted a search in Thomson Reuters
Web of Knowledge with the search term “fMRI OR functional magnetic resonance imaging
OR PET OR Positron Emission Tomography AND semantic AND sentence.” This search
yielded 409 studies, of which 66 reported contrasts between semantically more demanding and
less demanding sentences, contrasts between syntactically more-demanding and less-demanding
sentences, or contrasts between sentences and nonsentential control conditions. All these studies
tested healthy participants. The activation foci and the spatial extent of 198 contrasts were coded
in an anatomical reference system of 112 regions on the basis of the stereotaxic atlas of Talairach &
Tournoux (1988) (for details, see caption to Supplemental Table 2 and Indefrey & Levelt 2000,
2004).
For any particular region the reliability of its activation was assessed using the following esti-
mate: The average number of activated regions reported per experiment divided by the number
of regions (112) corresponds to the probability for any particular region to be reported in an
experiment, if reports were randomly distributed over regions. Assuming this probability, the
chance level for a region to be reported as activated in a certain number of experiments is given
by a binomial distribution. The possibility that the agreement of reports about a certain region
was coincidental was rejected if the chance level was below 5% (uncorrected for the number of
regions). Regions with a chance level below 0.0004 survived a Bonferroni correction for 112 re-
gions and are reported as 0.05 (corrected). This estimate takes into account that not all studies
covered the whole brain owing to the heterogeneity of techniques and analysis procedures (for
example, analyzing only regions of interest). The procedure also controls for the fact that the
average number of activated regions per study differs between contrasts. In contrasts comparing
sentences to low-level control conditions, the number of activated regions is typically higher than
in contrasts comparing syntactically demanding to less demanding sentences; thus the chances of

www.annualreviews.org • Language Beyond Single Words 351

NE37CH18-Hagoort ARI 30 June 2014 9:10

Supplemental Material
coincidental agreements between studies are also higher. Supplemental Tables 2, 3, and 4 show
the summary data for all 112 regions and all comparisons reported here.

Results
The contrasts we analyzed fall into two main categories: About one-third of the contrasts compared
sentences to nonsentential stimuli, ranging from word lists to cross-hair fixation or rest conditions.
The resulting brain activations can be expected to include whichever brain regions are involved in
Downloaded from www.annualreviews.org. ONOS (39152) IP: 202.41.10.108 On: Tue, 22 Jul 2025 11:51:37

sentence-level syntactic and semantic processing. However, many other regions subserving lower-
level processes, such as lexical retrieval, may also show up. About two-thirds of the contrasts com-
pared syntactically or semantically demanding sentences to less demanding sentences. The latter
studies controlled much more tightly for lower-level (e.g., lexical) differences between stimuli so
that the resulting activations could be considered specific to syntactic or semantic unification. Note,
however, that these studies not only may have missed neural correlates of sentence-level processing
that were shared between demanding and less demanding sentences, but also may have induced
processes related to higher general cognitive demands, such as attention or error-related processes.
The main manipulations for increasing syntactic demands are the use of sentences containing
syntactic violations or word-class ambiguities (e.g., “watch” as noun or verb) and the use of struc-
turally more complex sentences, such as those containing center-embedded object-relative clauses.
Manipulations for increasing semantic demands are semantic violations (“The trains were very
sour . . . ”) and lexical-semantic ambiguities (e.g., “bank”) that did not affect the syntactic struc-
ture in comparing with the correct control sentences. In addition, as instances of higher semantic
demands, we also considered experimental manipulations that complicated the listener’s ability
to assign an overall meaning without inducing a syntactic difference. These instances include
sentences with a metaphoric meaning, sentences inducing semantic operations such as coercion,
metonymy, and sentences making connections to the previous discourse context. These instances
also include sentences requiring listeners to assess speakers’ intentions (irony, indirect replies, or
requests; see also the section on neuropragmatics).

Sentences compared with control conditions below sentence level. Compared with control
conditions below sentence level, the comprehension of sentences reliably activates the temporal
lobes and the posterior IFG bilaterally (Figure 1a, Supplemental Figure 2, Supplemental
Table 2), albeit with a clear left hemisphere dominance. Not activated in any of the 53 studies
were large parts of the right parietal and right inferior temporal cortices. These areas are thus
reliably not involved during sentence comprehension. As shown in Supplemental Figure 3, some
differences can be seen among the regions involved in processing written and spoken sentences.
Some right hemisphere temporal regions are not reliably found in reading, and posterior frontal
regions are less frequently found in listening. Of particular interest are the results for a subset of
studies that used the most natural kind of sentence processing: The participants just listened or read

−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−→
Figure 1
Schematic representation of the brain showing regions with reliably reported activations for sentences compared with nonsentential
stimuli (a) and sentences with high syntactic or semantic processing demands compared with simpler sentences (b,c). The left posterior
inferior frontal gyrus is further subdivided into Brodmann areas (BA) 44 (above black line), BA 45 (below black line, above AC–PC line) and
BA 47 (below AC–PC line). Green regions indicate a reliable number of reports. Pink regions indicate no reports in 53 studies. For
details, see Supplemental Tables 2, 3, and 4 (follow the Supplemental Material link from the Annual Reviews home page at
http://www.annualreviews.org). Abbreviations: AC, anterior commissure; PC, posterior commissure).

352 Hagoort · Indefrey

NE37CH18-Hagoort ARI 30 June 2014 9:10

a Sentences compared with control conditions below sentence level

All (53 studies)
Downloaded from www.annualreviews.org. ONOS (39152) IP: 202.41.10.108 On: Tue, 22 Jul 2025 11:51:37

L R

Passive reading (13 studies) Passive listening (20 studies)

L R L R

b Sentences with higher compared with sentences with lower processing demands
Higher syntactic demands (57 studies)

L R

Higher semantic demands (51 studies)

L R
Anterior insula

c Direct comparisons between sentences with high syntactic and high semantic demands
Syntactic–semantic (6 studies) Semantic–syntactic (10 studies)

L L

Reliably activated p < 0.05, corrected Reliably activated p < 0.05, uncorrected Reliably not activated p < 0.05, uncorrected

www.annualreviews.org • Language Beyond Single Words 353

NE37CH18-Hagoort ARI 30 June 2014 9:10

Supplemental Material for comprehension without performing any additional tasks (Figure 1a, Supplemental Figure 4).
Irrespective of the input modality, passive sentence comprehension activates the left MTG, the
superior temporal gyrus (STG), and the orbital (BA 47) and triangular (BA 45) parts of LIFG. The
most dorsal part of the IFG (pars opercularis, BA 44), however, is not reliably activated during
passive listening to (simple) sentences. The same holds when sentence processing is compared
with the processing of word lists (see Supplemental Figure 2). This result suggests that either
BA 44 is not involved in any sentence-level unification process or this unification process is not
necessarily active for passive listening, for instance because a good-enough processing strategy
Downloaded from www.annualreviews.org. ONOS (39152) IP: 202.41.10.108 On: Tue, 22 Jul 2025 11:51:37

(Ferreira et al. 2002) does not require a full compositional analysis.

Sentences with higher demands on syntactic or semantic processing. So far, we have iden-
tified a set of candidate regions that may be involved in sentence-level processing. However, the
contrast of sentences compared with nonsentential stimuli neither excludes the involvement of
these regions in lexical processes nor allows for a distinction between semantic and syntactic uni-
fication. This limitation is not present for studies comparing syntactically or semantically more
demanding sentences with simpler sentences. In these studies, the words used in both conditions
are typically well-matched. A contribution of word-level differences to the resulting brain acti-
vation can hence be excluded. Despite that syntactic complexity may always have repercussions
for the construction of sentence meaning, semantic and syntactic processes can, in principle, be
independently taxed in this type of design. Figure 1b shows the results of 57 studies comparing
syntactically demanding sentences with less-demanding control sentences and 51 studies compar-
ing semantically demanding sentences with less-demanding control sentences. Higher syntactic
processing demands most reliably activate the more dorsal parts of posterior LIFG (BA 44/45),
the right posterior IFG and the left posterior STG and MTG. In addition, the left precuneus,
the left inferior parietal lobule, and the right posterior MTG all activated reliably. Higher se-
mantic processing demands most reliably activate all parts of posterior LIFG (but BA 45/47 are
reported twice as often as is BA 44; see Supplemental Figure 5, Supplemental Table 4), the right
posterior IFG, and the left middle and posterior MTG. In addition, the data indicate a reliable
activation of the medial prefrontal cortex that is not seen for higher syntactic processing demands
and demonstrate activations of the left anterior insula, angular gyrus, and the posterior ITG.
The results of 16 studies directly comparing sentences with high syntactic and high semantic
processing demands (Figure 1c, Supplemental Figure 6) confirm that the medial prefrontal
cortex is important for processing sentences with high semantic processing demands. Direct com-
parisons also demonstrate a syntactic/semantic gradient in the LIFG: a reliably stronger activation
of BA 44 is seen for syntactically, compared with semantically, demanding sentences; a reliably
stronger activation of BA 45/47 is observed for semantically, compared with syntactically, de-
manding sentences.
In sum, compared with studies contrasting relatively undemanding sentences with nonsenten-
tial control conditions, studies contrasting higher with lower demands on syntactic or semantic
sentence-level processing have found that IFG and some additional regions activate reliably.
However, they have found no reliable activation of the anterior temporal lobe, despite some claims
that it might be involved in semantic and syntactic combinatorial processing (Hickok & Poeppel
2007, Friederici 2012). We now consider the different kinds of experimental manipulations
that induced increased processing demands to determine in which way they contributed to the
observed overall results.

Violations, ambiguity, and complexity. Studies comparing sentences with syntactic viola-
tions (mostly agreement violations and phrase-structure/word-category violations) with correct

354 Hagoort · Indefrey

NE37CH18-Hagoort ARI 30 June 2014 9:10

sentences have most reliably found BA 44/45 activation (Supplemental Figure 7). Studies com- Supplemental Material
paring sentences containing semantic violations with correct sentences have most reliably found
activation of all parts of the left posterior IFG; activation of BA 45/47 was again reported more
often than was activation of BA 44. Both semantic and syntactic violations generally activate the
posterior temporal cortex less frequently than do demanding sentences. Within semantic vio-
lations, we also compared violations of semantic selection restrictions (e.g., “Dutch trains are
sour”/“My mother ironed a kiss”) with violations of world knowledge (e.g., “Dutch trains are
white”/“The woman painted the insect”; examples taken from Hagoort et al. 2004 and Kuperberg
Downloaded from www.annualreviews.org. ONOS (39152) IP: 202.41.10.108 On: Tue, 22 Jul 2025 11:51:37

et al. 2000). Both types of semantic violation activate BA 45/47 (Supplemental Figure 8). World
knowledge violations also seem to activate widespread medial brain structures.
Supplemental Figure 9 shows the reliably activated regions for sentences containing local syn-
tactic ambiguities (mostly word-class ambiguities, e.g., “He noticed that landing planes frightens
some new pilots”) or semantic ambiguities (“The reporter commented that modern compounds
react unpredictably”; examples from Rodd et al. 2010). Both types of ambiguities activate the
posterior IFG bilaterally and the left posterior MTG. For syntactic ambiguities, activation in the
left posterior IFG is confined to BA 44. Semantic ambiguities activate the left posterior inferior
medial temporal lobe.
Sixty-three studies increased the degree of syntactic or semantic complexity using means other
than violations or ambiguities. The majority of studies on syntactic complexity compared sentences
containing complex relative clauses with simpler relative clauses. The main manipulation in the
remaining studies was the use of noncanonical word order. Our criterion for studies inducing
semantic complexity was a comparison of a condition in which understanding the meaning of the
sentence required some additional effort compared with that required for syntactically identical
control sentences. In most of the studies this goal was achieved by comparing sentences containing
a metaphoric meaning (e.g., “A sailboat is a floating leaf ”; example from Diaz & Hogstrom 2011)
with sentences containing a literal meaning. Whereas a metaphor can be understood without
knowing the context of the utterance or the speaker, the same is not true for a different kind of
nonliteral sentence. To understand that ironic/sarcastic sentences or indirect replies/requests go
beyond a literal interpretation of the sentence requires that a listener assesses the situation and
the speaker’s intention (see below).
Both syntactic and semantic complexity reliably induce stronger activation of the posterior IFG
bilaterally and the left mid and posterior MTG (see Supplemental Figure 10). Left posterior IFG
activation again shows a gradient with activation of BA 44 for syntactic but not semantic complexity
and shows activation of BA 47 for semantic but not syntactic complexity. The posterior STG may
selectively be recruited for syntactic complexity. Conversely, semantic complexity induces medial
prefrontal activations that are not reliably seen for syntactic complexity manipulations.
Separate analyses of the two main kinds of syntactic complexity studied by investigators
(Supplemental Figure 11) yielded results that were similar to the overall activation patterns in-
duced by syntactic complexity; therefore, the mechanism that drives these activations seems to be
shared by noncanonical word orders and relative clause complexity. Separate analyses of different
kinds of semantic complexity, however, yielded differential activation patterns (Supplemental
Figure 12). Sentences with metaphoric meaning contributed most to the overall activation of
BA 45/47 and left posterior MTG. This result replicates the findings of a recent voxel-based
meta-analysis on metaphor processing (Bohrn et al. 2012; see contrast “metaphor > literal” in their
table 3). A few studies also reported activations of BA 47 and left MTG by comparing literal
sentences inducing additional semantic operations (metonymy, coercion, establishment of a
causal relationship with preceding context) with literal sentences without such operations. By
contrast, sentences that required the listener to assess the speaker’s intentions did not reliably

www.annualreviews.org • Language Beyond Single Words 355

NE37CH18-Hagoort ARI 30 June 2014 9:10

activate BA 45 or the left posterior temporal lobe. These kinds of sentences most frequently
activated the medial prefrontal cortex (also reliably reported for metaphoric sentences but in a
relatively smaller number of studies) and also the right temporoparietal cortex (mainly observed
in studies using indirect utterances).

Summary and Discussion

Our meta-analysis comparing reliable activations for the processing of various syntactically and
Downloaded from www.annualreviews.org. ONOS (39152) IP: 202.41.10.108 On: Tue, 22 Jul 2025 11:51:37

semantically demanding sentences yielded several important results. The most robust result is
a distinctive activation pattern in the posterior LIFG; syntactic demands activated more dorsal
parts (BA 44/45) and semantic demands activated more ventral parts (BA 45/47) across all kinds
of increased processing demands (violations, ambiguity, complexity) and in studies performing
direct comparisons of high syntactic and semantic processing demands (see Figure 2a). This result
indicates that syntactic unification cannot simply be reduced to semantic unification. In particular,
BA 44 activation is clearly driven more strongly by syntactic than by semantic demands, suggesting
that this region contains neuronal populations involved in syntactic operations as such or that the
semantic consequences of syntactic demands (difficulty of thematic role assignment) are processed
by neuronal populations that differ from those processing other kinds of semantic unification.
A novel observation is that the dorsal/ventral gradient observed in the left posterior IFG seems
to be mirrored in the left posterior temporal lobe. Higher syntactic demands reliably activate the
STG and MTG, and higher semantic demands reliably activate the MTG and ITG.
To confirm these gradients, we conducted a spatially more fine-grained analysis on a subset of
28 studies reporting Talairach or MNI coordinates of activation foci in the left posterior temporal
cortex. Whereas the mean locations of the activations reported for syntactically (mean y = −45.0,
SD = 7.7) and semantically (mean y = −46.5, SD = 5.9) demanding sentences did not differ
significantly in the rostral/caudal dimension, the mean location of activation foci for semantically
demanding sentences was significantly more ventral (middle temporal gyrus, mean z = 0.9, SD =
8.6) than the mean location of activation foci for syntactically demanding sentences (superior
Supplemental Material temporal sulcus, mean z = 10.2, SD = 5.7; see Figure 2b and Supplemental Figure 13).
Activations in the left posterior inferior frontal gyrus reported for the same contrasts in this set
of studies confirmed a significantly more ventral and rostral mean location of activation foci for
semantically demanding sentences (BA 45, mean y = 22.2, SD y = 6.3, mean z = 4.2, SD z =
10.8) as compared with syntactically demanding sentences (BA 44, mean y = 14.8, SD y = 8.0,
mean z = 15.0, SD z = 5.6).
These corresponding gradients in posterior frontal and temporal regions are remarkably con-
sistent with a functional connectivity pattern found by Xiang et al. (2010), which links seed regions
in BA 44, BA 45, and BA 45/47 to left posterior STG, MTG, and ITG, respectively. This finding
−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−−→
Figure 2
(a) Summary of activation patterns for sentences with high syntactic or semantic processing demands
compared with simpler sentences. (b) Syntactic/semantic gradients in left inferior frontal and posterior
temporal cortex based on 28 studies reporting posterior temporal cortex activation for syntactically
demanding or semantically demanding sentences compared with less demanding sentences (see
Supplemental Figure 13 for details). The centers represent the mean coordinates of the local maxima, and
the radii represent the standard deviations of the distance between the local maxima and their means.
Abbreviations: GFm, GFi, middle and inferior frontal gyri; BA, Brodmann area; GTs, GTm, GTi, superior,
middle, and inferior temporal gyri; STS, ITS, superior and inferior temporal sulci; Gsm, supramarginal
gyrus.

356 Hagoort · Indefrey

NE37CH18-Hagoort ARI 30 June 2014 9:10

clearly supports the idea that sentence-level unification relies on the coactivation of neuronal pop-
ulations in a network of posterior frontal and temporal regions, with a similar functional gradient
in both parts of the brain.
Another important observation is the degree to which posterior temporal lobe activation dif-
fers between violations and other kinds of higher processing demands. Syntactic violations do

a
Downloaded from www.annualreviews.org. ONOS (39152) IP: 202.41.10.108 On: Tue, 22 Jul 2025 11:51:37

L L

Activated by all high syntactic Activated by all high semantic demands

demands, more strongly activated by (except irony) and more strongly activated
high syntactic versus semantic demands by high semantic versus syntactic demands
Activated by all high syntactic demands Activated by all nonliteral sentences
(except ambiguity), less strongly activated (in particular speaker meaning) and
by high syntactic versus semantic demands semantic violations, more strongly activated
by high semantic versus syntactic demands
Activated by all high syntactic demands
(except violation), less strongly activated Activated by semantic ambiguity
by high syntactic versus semantic demands

Activated by syntactically complex

sentences but not ambiguity and violation

b 50

40

30 G
m
GF s
m
BA44
20
STS
SY
mm

10 m
i BA45 GT
GF
0
s
BA47 GT GT
i
m
–10 GT
STS
ITS
–20
i
GT Semantically demanding
–30 Syntactically demanding

40 30 20 10 0 –10 –20 –30 –40 –50 –60 –70

mm

www.annualreviews.org • Language Beyond Single Words 357

NE37CH18-Hagoort ARI 30 June 2014 9:10

not seem to elicit posterior temporal lobe activations reliably, and reports about such activations
are relatively infrequent for semantic violations. A plausible account for this dissociation may be
based on a distinction between Broca’s area subserving sentence-level compositional processes
and the posterior temporal lobe subserving the retrieval of lexical syntactic and semantic informa-
tion (Hagoort 2005). Violation paradigms may tap relatively specifically into a unification stage,
whereas ambiguities and complexity manipulations typically also require lexical syntactic and se-
mantic specifications to be retrieved. In sentences with unambiguous violations (“The blouse was
on ironed”/“The thunderstorm was ironed”; examples from Friederici et al. 2003), no alternative
Downloaded from www.annualreviews.org. ONOS (39152) IP: 202.41.10.108 On: Tue, 22 Jul 2025 11:51:37

lexical specifications can be constructed to help remedy the problem; however, not only for am-
biguities (“he noticed that landing planes frightens some new pilots”/“the reporter commented
that modern compounds react unpredictably”), but also for many kinds of syntactic and semantic
complexities, alternative lexical specifications may solve the comprehension problem.
Finally, our results suggest that the neuronal substrate underlying the processing of ironic
and indirect utterances differs from that activated in other kinds of high-demand sentences. Such
utterances do not seem to induce reliable activation of Broca’s area (BA 44/45) and Wernicke’s
area but, instead, show the most reliable activation in the medial prefrontal cortex and the right
temporoparietal cortex. The difficulty in processing these sentences does not seem to be in de-
coding the meaning of the sentence but in a different aspect of comprehension to which we now
turn in more detail.

BEYOND CORE AREAS FOR LANGUAGE: THE CASE

OF NEUROPRAGMATICS
In many instances, linguistic expressions are underdetermined with respect to the meaning they
convey. What is said and what is understood are often not the same. Communication goes further
than the exchange of explicit propositions. In essence, the goal of the speaker is either to change
the listener’s mind or to commit the addressee to the execution of certain actions, such as closing
the window in reply to the statement, “It is cold here.” A theory of speech acts is needed to
understand how we move from coded meaning to inferred speaker meaning (cf. Grice 1989,
Levinson 2013, Hagoort & Levinson 2015). The steps required to progress from coded meaning
to speaker meaning involve additional neuronal systems, as recent studies have shown.
In a study on conversational implicatures and indirect requests, for example, Bašnáková and
coworkers (2013) contrasted direct and indirect replies, two classes of utterances whose speaker
meanings are either very similar to, or markedly different from, their coded meaning. In this
study, participants listened to natural spoken dialogue in which the final utterance, e.g., “It is
hard to give a good presentation,” had different meanings depending on the dialogue context and
the immediately preceding question. This final utterance either served as a direct reply (to the
question “How hard is it to give a good presentation?”) or as an indirect reply (to “Did you like my
presentation?”). In the indirect reply condition, activations were observed in the medial prefrontal
cortex (mPFC) extending into the right anterior part of the supplementary motor area (SMA),
and in the right temporoparietal junction (TPJ). The data thus demonstrated a pattern typical for
tasks that involve mentalizing based on a theory of mind (ToM) (Amodio & Frith 2006, Mitchell
et al. 2006, Saxe et al. 2006). Although the exact role of all the individual ToM regions is not yet
clearly established, both mPFC and right TPJ constitute core regions in ToM tasks (Carrington
& Bailey 2009). The most specific hypothesis about the role of the posterior part of right TPJ
(Mars et al. 2012) in the mentalizing network is that it is implicated in mental state reasoning, i.e.,
thinking about other people’s beliefs, emotions, and desires (Saxe 2010).

358 Hagoort · Indefrey

NE37CH18-Hagoort ARI 30 June 2014 9:10

mPFC activation in the Bašnáková et al. study (2013) was found in parts of the mPFC associ-
ated with complex sociocognitive processes such as mentalizing or thinking about the intentions of
others (such as communicative intentions, right anterior mPFC) or about oneself (right posterior
mPFC). The involvement of these regions is also consistently observed in discourse comprehension
(e.g., Mason & Just 2009, Mar 2011). This finding might come as no surprise because the mo-
tivations, goals, and desires of fictional characters are likely accessed in a manner similar to that
of real-life protagonists (Mar & Oatley 2008). In fact, an influential model from the discourse-
processing literature (Mason & Just 2009) ascribes to the dorsomedial part of the frontal cortex and
Downloaded from www.annualreviews.org. ONOS (39152) IP: 202.41.10.108 On: Tue, 22 Jul 2025 11:51:37

the right TPJ a functional role as a protagonist perspective network, which generates expectations
about how the protagonists of stories will act on the basis of understanding their intentions.
Although the literature on the neuropragmatics of language is still limited, studies have, with
remarkable consistency, demonstrated that understanding the communicative intent of an ut-
terance requires mentalizing. Because the linguistic code underdetermines speaker meaning, the
ToM network needs to be invoked to progress from coded meaning to speaker meaning. Despite
the popular view that the mirror neuron system (MNS) is sufficient for action understanding
(Rizzolatti & Sinigaglia 2010), the MNS does not provide the crucial neural infrastructure for in-
ferring speaker meaning. In addition to core areas for retrieving lexical information from memory
and unifying the lexical building blocks to produce and understand multiword utterances, other
brain networks are needed to realize language-driven communication to its full extent (for a more
extended review of studies on neuropragmatics, see Hagoort & Levinson 2015).

CONCLUSION
We have outlined the contours of the neurobiological infrastructure that supports language pro-
cessing beyond the comprehension or production of single words. In this context, research has
established substantial deviations of the classical WLG model, with its focus on single-word pro-
cessing. Three major deviations are worth highlighting: (a) Although not discussed in detail here
(for more extended discussions, see Hagoort 2013, Amunts & Catani 2015) the connectivity of
the language cortex in left perisylvian regions is much more extended than was proposed in the
classical model and is certainly not restricted to the arcuate fasciculus; (b) the distribution of labor
between the core regions in left perisylvian cortex is fundamentally different than that proposed
in the classical model (in contrast to the claims of the WGL model, there is strong evidence
for shared circuitry for core aspects of sentence-level language production and comprehension);
and (c) the operation of language to its full extent requires a much more extended network than
what the classical model assumed. The basic principle of brain organization for higher cognitive
functions proposes that these functions are based on the interaction between numerous neuronal
circuits and brain regions that support the various contributing functional components. These
circuits are not necessarily specialized for language but nevertheless need to be recruited for the
sake of successful language processing. One example is the ToM network, which seems critical
for speakers to construct utterances with knowledge of the listener in mind and for listeners to
progress from coded meaning to speaker meaning. As described above in the section on the tem-
poral profile of unification, the network of language areas in the brain is highly dynamic despite
the static nature suggested by pictures of the neuronal infrastructure for language. The specific
contribution that any area makes to information processing depends on the input it receives at a
certain time step, which itself depends on the computational environment in which it is embedded.
Moreover, a large meta-analysis has at last confirmed a clear division of labor in both frontal and
temporal areas for syntactic and semantic unification.

www.annualreviews.org • Language Beyond Single Words 359

NE37CH18-Hagoort ARI 30 June 2014 9:10

DISCLOSURE STATEMENT
The authors are not aware of any affiliations, memberships, funding, or financial holdings that
might be perceived as affecting the objectivity of this review.

ACKNOWLEDGMENTS
This research was in part supported by a Spinoza Award to P.H. and by support to P.I. from the
Collaborative Research Center 991 of the Deutsche Forschungsgemeinschaft (DFG). We thank
Downloaded from www.annualreviews.org. ONOS (39152) IP: 202.41.10.108 On: Tue, 22 Jul 2025 11:51:37

Frauke Hellwig for her assistance with the artwork.

LITERATURE CITED
Amodio DM, Frith CD. 2006. Meeting of minds: the medial frontal cortex and social cognition. Nat. Rev.
Neurosci. 7:268–77
Amunts K, Catani M. 2015. Cytoarchitectonics, receptor architectonics and network topology of language.
See Gazzaniga 2015. In press
Bašnáková J, Weber K, Petersson KM, van Berkum J, Hagoort P. 2013. Beyond the language given: the neural
correlates of inferring speaker meaning. Cereb. Cortex. In press
Bohrn IC, Altmann U, Jacobs AM. 2012. Looking at the brains behind figurative language—a quantita-
tive meta-analysis of neuroimaging studies on metaphor, idiom, and irony processing. Neuropsychologia
50:2669–83
Caramazza A, Zurif EB. 1976. Dissociation of algorithmic and heuristic processes in language comprehension:
evidence from aphasia. Brain Lang. 3:572–82
Carrington SJ, Bailey AJ. 2009. Are there theory of mind regions in the brain? A review of the neuroimaging
literature. Hum. Brain Mapp. 30:2313–35
Chen E, West WC, Waters G, Caplan D. 2006. Determinants of bold signal correlates of processing object-
extracted relative clauses. Cortex 42:591–604
DeLong KA, Urbach TP, Kutas M. 2005. Probabilistic word pre-activation during language comprehension
inferred from electrical brain activity. Nat. Neurosci. 8:1117–21
Diaz MT, Hogstrom LJ. 2011. The influence of context on hemispheric recruitment during metaphor pro-
cessing. J. Cogn. Neurosci. 23:3586–97
Ferreira F, Bailey KGD, Ferraro V. 2002. Good-enough representations in language comprehension. Curr.
Dir. Psychol. Sci. 11:11–15
Friederici AD. 2012. The cortical language circuit: from auditory perception to sentence comprehension.
Trends Cogn. Sci. 16:262–68
Friederici AD, Rüschemeyer SA, Hahne A, Fiebach CJ. 2003. The role of left inferior frontal and superior
temporal cortex in sentence comprehension: localizing syntactic and semantic processes. Cereb. Cortex
13:170–77
Garrido MI, Kilner JM, Kiebel SJ, Friston KJ. 2007. Evoked brain responses are generated by feedback loops.
Proc. Natl. Acad. Sci. USA 104:20961–66
Gazzaniga MS, ed. 2015. The Cognitive Neurosciences. Cambridge, MA: MIT Press. 5th ed. In press
Grainger J, Rey A, Dufau S. 2008. Letter perception: from pixels to pandemonium. Trends Cogn. Sci. 12:381–87
Grice P. 1989. Studies in the Way of Words. Cambridge, MA: Harvard Univ. Press
Grill-Spector K, Henson R, Martin A. 2006. Repetition and the brain: neural models of stimulus-specific
effects. Trends Cogn. Sci. 10:14–23
Hagoort P. 2005. On Broca, brain, and binding: a new framework. Trends Cogn. Sci. 9:416–23
Hagoort P. 2013. MUC (memory, unification, control) and beyond. Front. Psychol. 4:416
Hagoort P, Hald L, Bastiaansen M, Petersson KM. 2004. Integration of word meaning and world knowledge
in language comprehension. Science 304:438–41
Hagoort P, Levinson SL. 2015. Neuropragmatics. See Gazzaniga 2015. In press
Hagoort P, van Berkum J. 2007. Beyond the sentence given. Philos. Trans. R. Soc. B 362:801–11

360 Hagoort · Indefrey

NE37CH18-Hagoort ARI 30 June 2014 9:10

Hickok G, Poeppel D. 2007. The cortical organization of speech processing. Nat. Rev. Neurosci. 8:393–402
Indefrey P. 2011. Neurobiology of syntax. In The Cambridge Encyclopedia of the Language Sciences, ed.
PC Hogan, pp. 835–38. Cambridge, UK/New York: Cambridge Univ. Press
Indefrey P. 2012. Hemodynamic studies of syntactic processing. In The Handbook of the Neuropsychology of
Language, ed. M Faust, pp. 209–28. Malden, MA: Blackwell
Indefrey P, Hellwig F, Herzog H, Seitz RJ, Hagoort P. 2004. Neural responses to the production and com-
prehension of syntax in identical utterances. Brain Lang. 89:312–19
Indefrey P, Levelt WJM. 2000. The neural correlates of language production. In The New Cognitive Neuro-
sciences, ed. MS Gazzaniga, pp. 845–65. Cambridge, MA: MIT Press
Downloaded from www.annualreviews.org. ONOS (39152) IP: 202.41.10.108 On: Tue, 22 Jul 2025 11:51:37

Indefrey P, Levelt WJM. 2004. The spatial and temporal signatures of word production components. Cognition
92:101–44
Jackendoff R. 2002. Foundations of Language: Brain, Meaning, Grammar, Evolution. Oxford, UK: Oxford Univ.
Press
Jackendoff R. 2007. A parallel architecture perspective on language processing. Brain Res. 1146:2–22
Kuperberg GR, McGuire PK, Bullmore ET, Brammer MJ, Rabe-Hesketh S, et al. 2000. Common and distinct
neural substrates for pragmatic, semantic, and syntactic processing of spoken sentences: an fMRI study.
J. Cogn. Neurosci. 12:321–41
Kuperberg GR, Sitnikova T, Lakshmanan BM. 2008. Neuroanatomical distinctions within the semantic system
during sentence comprehension: evidence from functional magnetic resonance imaging. NeuroImage
40:367–88
Kutas M, Federmeier KD, Urbach ThP. 2015. The “negatives” and “positives” of prediction in language. See
Gazzaniga 2015. In press
Lau E, Stroud C, Plesch S, Phillips C. 2006. The role of structural prediction in rapid syntactic analysis. Brain
Lang. 98:74–88
Levelt WJM. 2013. A History of Psycholinguistics: The Pre-Chomskyan Era. Oxford, UK: Oxford Univ. Press
Levinson SC. 2013. Action formation and ascription. In The Handbook of Conversation Analysis, ed. J Sidnell,
T Stivers, pp. 103–30. Malden, MA: Wiley-Blackwell
Mar RA. 2011. The neural bases of social cognition and story comprehension. Annu. Rev. Psychol. 62:103–34
Mar RA, Oatley K. 2008. The function of fiction is the abstraction and simulation of social experience. Perspect.
Psychol. Sci. 3:173–92
Mars RB, Sallet J, Schüffelgen U, Jbabdi S, Toni I, Rushworth MF. 2012. Connectivity-based subdivisions of
the human right “temporoparietal junction area”: evidence for different areas participating in different
cortical networks. Cereb. Cortex 22:1894–903
Mason RA, Just MA. 2009. The role of the theory-of-mind cortical network in the comprehension of narratives.
Lang. Linguist. Compass 3:157–74
Menenti L, Gierhan SME, Segaert K, Hagoort P. 2011. Shared language: overlap and segregation of the
neuronal infrastructure for speaking and listening revealed by functional MRI. Psychol. Sci. 22:1173–82
Metusalem R, Kutas M, Urbach TP, Hare M, McRae K, Elman JL. 2012. Generalized event knowledge
activation during online sentence comprehension. J. Mem. Lang. 66:545–67
Meyer AS, Hagoort P. 2013. What does it mean to predict one’s own utterances? Behav. Brain Sci. 36:367–68
Mitchell JP, Macrae CN, Banaji MR. 2006. Dissociable medial prefrontal contributions to judgments of
similar and dissimilar others. Neuron 50:655–63
Rizzolatti G, Sinigaglia C. 2010. The functional role of the parieto-frontal mirror circuit: interpretations and
misinterpretations. Nat. Rev. Neurosci. 11:264–74
Rodd JM, Longe OA, Randall B, Tyler LK. 2010. The functional organisation of the fronto-temporal language
system: evidence from syntactic and semantic ambiguity. Neuropsychologia 48:1324–35
Saxe R, Moran JM, Scholz J, Gabrieli J. 2006. Overlapping and non-overlapping brain regions for theory of
mind and self reflection in individual subjects. Soc. Cogn. Affect. Neurosci. 1:229–34
Saxe, R. 2010. The right temporo-parietal junction: a specific brain region for thinking about thoughts. In
Handbook of Theory of Mind, ed. A Leslie, T German, pp. 1–35. Hove, UK: Psychology Press
Segaert K, Menenti L, Weber K, Petersson KM, Hagoort P. 2012. Shared syntax in language production and
language comprehension—an FMRI study. Cereb. Cortex 22:1662–70

www.annualreviews.org • Language Beyond Single Words 361

NE37CH18-Hagoort ARI 30 June 2014 9:10

Steinhauer K, Drury JE. 2012. On the early left-anterior negativity (ELAN) in syntax studies. Brain Lang.
120:135–62
Szewczyk JM, Schriefers H. 2013. Prediction in language comprehension beyond specific words: an ERP study
on sentence comprehension in Polish. J. Mem. Lang. 68:297–314
Talairach J, Tournoux P. 1988. Co-Planar Stereotaxic Atlas of the Human Brain: 3-D Proportional System: An
Approach to Cerebral Imaging. Stuttgart: Thieme
Thorpe S, Fize D, Marlot C. 1996. Speed of processing in the human visual system. Nature 381:520–22
Tyler LK, Cheung TP, Devereux BJ, Clarke A. 2013. Syntactic computations in the language network:
characterizing dynamic network properties using representational similarity analysis. Front. Psychol. 4:271
Downloaded from www.annualreviews.org. ONOS (39152) IP: 202.41.10.108 On: Tue, 22 Jul 2025 11:51:37

Tyler LK, Marslen-Wilson WD, Randall B, Wright P, Devereux BJ, et al. 2011. Left inferior frontal cortex
and syntax: function, structure and behaviour in patients with left hemisphere damage. Brain 134:415–31
Van Berkum JJ, Brown CM, Zwitserlood P, Kooijman V, Hagoort P. 2005. Anticipating upcoming words in
discourse: evidence from ERPs and reading times. J. Exp. Psychol. Learn. Mem. Cogn. 31:443–67
Van Berkum JJA, van den Brink D, Tesink C, Kos M, Hagoort P. 2008. The neural integration of speaker
and message. J. Cogn. Neurosci. 20:580–91
Von Humboldt W. 1829. On the dual. Nouv. Rev. Ger. 1:378–81
Willems RM, Özyürek A, Hagoort P. 2007. When language meets action: the neural integration of gesture
and speech. Cereb. Cortex 17:2322–33
Xiang H-D, Fonteijn HM, Norris DG, Hagoort P. 2010. Topographical functional connectivity pattern in
the Perisylvian language networks. Cereb. Cortex 20:549–60
Zwitserlood P. 1989. The locus of the effects of sentential-semantic context in spoken-word processing.
Cognition 32:25–64

362 Hagoort · Indefrey