C H A P T E R

28
MUC (Memory, Unification, Control):
A Model on the Neurobiology of Language
Beyond Single Word Processing
Peter Hagoort
Donders Institute for Brain, Cognition and Behaviour, Max Planck Institute for Psycholinguistics,
Nijmegen, The Netherlands

28.1 INTRODUCTION knowledge about language processing beyond single

words (Hagoort, 2005, 2013). After describing its three
Until not too long ago, the neurobiological model components, I discuss the evidence that has accumu-
that has dominated our view on the neural architecture lated in support of the model.
of language was the Wernicke-Lichtheim-Geschwind
model. In this classical model, the human language
faculty was situated in the left perisylvian cortex, with 28.2 MEMORY, UNIFICATION,
a division of labor between the frontal and temporal AND CONTROL
regions. Wernicke’s area in left temporal cortex was
assumed to subserve the comprehension of speech, The MUC model distinguishes three functional com-
whereas Broca’s area in left inferior frontal cortex ponents of language processing: Memory, Unification,
(LIFC) was claimed to subserve language production. and Control. The Memory component refers to the lin-
The arcuate fasciculus connected these two areas. This guistic knowledge that in the course of language acqui-
model was based on single word processing. Since sition gets consolidated in neocortical memory
then, researchers interested in brain and language structures (see Davis & Gaskell, 2009, for the shift from
have realized that language is more than the concate- medial temporal lobe to neocortical structures during
nation of single words. Research focusing on sentence consolidation). It is the only language-specific compo-
processing has found that lesions in Broca’s region and nent of the model. The knowledge about the building
adjacent cortex impair not only language production blocks of language that is stored in memory (e.g., pho-
but also language comprehension (Caramazza & Zurif, nological, morphological, syntactic building blocks;
1976), whereas lesions in Wernicke’s region not only jointly referred to as lexical items) is domain-specific
affect language comprehension but also language pro- and, hence, coded in a format that is different from,
duction. More recent neuroimaging studies provided for example, color and visual object information.
further evidence that central aspects of language pro- However, language processing is more than mem-
duction and comprehension are subserved by shared ory retrieval and more than the simple concatenation
neural circuitry (Menenti, Gierhan, Segaert, & Hagoort, of retrieved lexical items. The expressive power of
2011; Segaert, Menenti, Weber, Petersson, & Hagoort, human language derives from being able to combine
2012). Since the advent of a whole toolkit of neuroim- elements from memory in novel ways. This process of
aging techniques, new models of the neural architec- deriving higher-level (i.e., sentence and beyond) mean-
ture of human language skills have been proposed. ing is referred to as Unification. Although as a result
Here, I focus mainly on the Memory-Unification- of the Chomskyan revolution in linguistics psycholin-
Control (MUC) model as a model that tries to integrate guistic studies of unification have mainly focused on

Neurobiology of Language. DOI: http://dx.doi.org/10.1016/B978-0-12-407794-2.00028-6 339 © 2016 Elsevier Inc. All rights reserved.
340 28. MUC (MEMORY, UNIFICATION, CONTROL): A MODEL ON THE NEUROBIOLOGY OF LANGUAGE BEYOND SINGLE WORD PROCESSING

syntactic analysis, unification operations not only take retrieved from memory. Within LIFC (Unification
place at the syntactic processing level but also are a Space), there seems to be a certain spatial distribution
hallmark of language across representational domains of recruitment dependent on the type of information
(Jackendoff, 2002). Thus, at the semantic and phono- that gets unified. Semantic unification recruits BA 47
logical levels, lexical elements are combined and inte- and BA 45; syntactic unification has its focus in BA 45
grated into larger structures. Hence, I distinguish and BA 44; phonological processes recruit BA 44 and
between syntactic, semantic, and phonological unifica- ventral parts of BA 6 (Hagoort & Indefrey, 2014). In
tion (Hagoort, 2005). addition, executive control needs to be exerted such
Finally, the Control component relates language to that the correct target language is selected, turn-taking
joint action and social interaction, and it is invoked, for in conversation is orchestrated, attention is given to
instance, when the contextually appropriate target lan- the most relevant information in the input, and so
guage has to be selected, or for handling the joint forth. Control regions involve dorsolateral prefrontal
action aspects of using language in conversational set- cortex (in pink) and midline structure, including the
tings. Later, it is shown how languages have built-in anterior cingulate cortex and the parts of parietal cor-
linguistic devices that trigger the attentional control tex that are involved in attention (not shown in
system into operation. Figure 28.1).
In the MUC model, the distribution of labor is as The distribution of labor in the MUC model is not
follows (Figure 28.1). Regions in the temporal cortex absolute. Language functions do not reside in single
(in yellow) and the angular gyrus in parietal cortex brain regions. Instead, language is subserved by
subserve the knowledge representations that have dynamic networks of brain regions, including the ones
been laid down in memory during acquisition. These outlined here. Ultimately, the mapping of a given lan-
regions store information, including phonological guage function onto the neural architecture of the
word forms, morphological information, word mean- brain is in terms of a network of brain areas instantiat-
ings, and the syntactic templates associated with noun, ing that particular language function (McIntosh, 2008;
verbs, and adjectives (for details, see Hagoort, 2003, Mesulam, 1998; Sporns, 2011). This is what Fedorenko
2005, 2009). Dependent on knowledge type, different and Thompson-Schill (2014) refer to as Networks of
parts of temporal cortex are involved. Frontal regions Interest. Typically, each node in such a network will
(Broca’s area and adjacent cortex; in blue) are crucial participate dynamically in other functional networks
for unification operations. These operations generate as well. Although one can claim a certain contribution
larger structures from the building blocks that are of a specific region (e.g., part of Broca’s area), it is cru-
cial to realize that such a contribution depends on the
interaction with other regions that are part of the net-
work. In short, “the mapping between neurons and
cognition relies less on what individual nodes can do
and more on the topology of their connectivity”
(Sporns, 2011, p. 184). Therefore, before discussing the
empirical evidence for the distribution of labor within
the MUC framework, I discuss the connectivity profile
of the language networks in the brain.

28.3 THE NETWORK TOPOLOGY OF THE

LANGUAGE CORTEX

The classical model has given the arcuate fasciculus

a central role in connecting the language-relevant parts
FIGURE 28.1 The MUC model of language. The figure displays of the brain. This was based on the idea that Broca’s
a lateral view of the left hemisphere. The numbers indicate area and Wernicke’s area were the two central nodes
Brodmann areas. These are areas with differences in the cytoarchitec- in the language network. The language network is
tonics (i.e., composition of cell types). The memory areas are in the much more extended than was assumed in the classi-
temporal cortex (in yellow) including the angular gyrus in parietal cal model and includes not only regions in the left
cortex. Unification requires the contribution of Broca’s area
(Brodmann areas 44 and 45) and adjacent cortex (Brodmann areas 47 hemisphere but also the right hemisphere areas.
and 6) in the frontal lobe. Control operations recruit another part of However, the evidence of additional activations in the
the frontal lobe (in pink) and the anterior cingulate cortex, as well as right hemisphere and areas other than Broca’s and
areas involved in attention (not shown in the figure). Wernicke’s does not take away the crucial role of left

D. LARGE-SCALE MODELS
 28.3 THE NETWORK TOPOLOGY OF THE LANGUAGE CORTEX 341

FIGURE 28.2 The arcuate fasciculus in a human, chimpanzee, and macaque in a schematic lateral view of the left hemisphere. From
Rilling et al. (2008), courtesy of Nature Publishing Group.

perisylvian cortex. In a recent meta-analysis based on

128 neuroimaging studies, Vigneau et al. (2010) com-
pared left and right hemisphere activations observed
in relation to language processing. For phonological
processing, lexico-semantic processing, and sentence
or text processing, the number of activation peaks in mc
AF ag
the right hemisphere comprised less than one-third of
the activation peaks in the left hemisphere. Moreover,
in the majority of cases the right hemisphere activa- op
tr ac vc
tions were found in homotopic regions, suggesting a tg
or
strong interhemispheric dependency. It is therefore EC
fg
justified to think that for the majority of the human
UF ILF
population (e.g., with the exception of some portion of
tp
left-handers, cases of left hemispherectomy), the
language-readiness of the human brain is strongly but
not exclusively based on the organization of the left FIGURE 28.3 Simplified illustration of the anatomy and connec-
perisylvian cortex. This, however, does not deny the tivity of the left hemisphere language network. Cortical areas are
relevant contributions of the right hemisphere in, for represented as red circles: pars orbitalis (or), pars triangularis (tr), and
instance, speech recognition (Hickok & Poeppel, 2007). pars opercularis (op) of the LIFC, angular gyrus (ag), superior and
middle temporal gyri (tg), fusiform gyrus (fg), and temporal pole (tp).
A recent technique for tracing fiber bundles in the liv-
ing brain is diffusion tensor imaging (DTI). Using DTI,
Rilling et al. (2008) tracked the arcuate fasciculus in
humans, chimpanzees, and macaques. These authors
found in humans a prominent temporal lobe projection progressive aphasia after damage to the dorsal tracts but
of the arcuate fasciculus that is much smaller or absent not after damage to the ventral tracts. This suggests that
in nonhuman primates (Figure 28.2). Moreover, connec- the dorsal tracts, including the arcuate fasciculus, are a
tivity with the middle temporal gyrus (MTG) was more key component in connecting frontal and temporal
widespread in the left than in the right hemisphere. This regions involved in syntactic processing. Again, exclu-
human specialization may be relevant for the evolution sivity is difficult to establish. Part of these tracts might
of language. Catani et al. (2007) found that the human also subserve other aspects of language processing.
arcuate fasciculus is strongly lateralized to the left, with In addition to the arcuate fasciculus, other fiber
quite some variation on the right. On the right, some bundles are important in connecting frontal with tem-
people lack an arcuate fasciculus, in others it is smaller poroparietal language regions (Figure 28.3). These
in size, and only in a minority of the population is this include the superior longitudinal fasciculus (adjacent
fiber bundle of equal size in both hemispheres. This pat- to the arcuate fasciculus) and the extreme capsule fas-
tern of lateralization was confirmed in a study involving ciculus, as well as the uncinate fasciculus, connecting
183 healthy right-handed volunteers in the age range Broca’s area with superior and middle temporal cortex
between 5 and 30 years (Lebel & Beaulieu, 2009). The along a ventral path (Anwander, Tittgemeyer, von
functionality of the arcuate fasciculus is not limited to Cramon, Friederici, & Knosche, 2007; Friederici, 2009;
single word processing. In a recent work, Wilson, Kelly et al., 2010). Figure 28.3 provides a schematic
Galantucci, Tartaglia, and Gorno-Tempini (2012) overview of the more extended connectivity profile of
reported syntactic deficits in patients with primary the left perisylvian cortex.

D. LARGE-SCALE MODELS
342 28. MUC (MEMORY, UNIFICATION, CONTROL): A MODEL ON THE NEUROBIOLOGY OF LANGUAGE BEYOND SINGLE WORD PROCESSING

of labor between Broca’s area, Wernicke’s area, and

adjacent regions. However, the proposed distribution
of labor is not absolute, but rather embedded and situ-
ated in the network skeleton of the language system’s
P1
neural architecture.
What is the evidence for the relative division of
P2
Oper
labor proposed in the MUC model? Let us consider
P3
the syntactic network first. In comparison with phono-
Tri
logical and semantic processing, which have compel-
Orbi T1
ling bilateral contributions, syntactic processing seems
T2 strongly lateralized to the left hemisphere perisylvian
T3
regions. Indirect support for a distinction between a
memory component (i.e., the mental lexicon) and a
unification component in syntactic processing comes
from neuroimaging studies on syntactic processing.
Two regions have been systematically reported in rela-
FIGURE 28.4 The topographical connectivity pattern between
tion to syntactic processing (Hagoort & Indefrey, 2014):
frontal and temporal/parietal cortex in the perisylvian language net-
works. Connections to the left pars opercularis (oper), pars triangu- the left posterior superior/middle temporal gyrus
laris (tri), and pars orbitalis (orbi) are shown in black, dark grey, and (STG/MTG) and the LIFC. The left posterior temporal
white arrows, respectively. The solid arrows represent the main cortex is known to be involved in lexical processing
(most significant) correlations and the dashed arrows represent the (Hickok & Poeppel, 2004, 2007; Indefrey & Cutler,
extending (overlapping) connections. Brain areas assumed to be
mainly involved in phonological, syntactic, and semantic processing
2004; Lau, Stroud, Plesch, & Phillips, 2006). In connec-
are shown in black, dark grey, and light grey circles, respectively. tion to the MUC model, this part of the brain might be
P1, supramarginal gyrus; P3, angular gyrus (AG); P2, the area important for the retrieval of the syntactic frames that
between SMG and AG in the superior/inferior parietal lobule; T1, are stored in the lexicon. The idea of syntactic frames
posterior superior temporal gyrus; T2, posterior MTG; P3, posterior that specify the possible local syntactic environment of
inferior temporal gyrus.
a given lexical item is in line with linguistic and
computational approaches that assume syntactic
knowledge to be lexically specified (Culicover &
Using resting state fMRI, Xiang, Fonteijn, Norris, and Jackendoff, 2006; Joshi & Schabes, 1997; Vosse &
Hagoort (2010) found a clear topographically organized Kempen, 2000). The Unification Space, where individ-
connectivity pattern in the left inferior frontal, parietal, ual frames are connected into a phrasal configuration
and temporal regions (Figure 28.4). In the left—but not for the whole utterance, might recruit the contribution
in the right—perisylvian cortex, functional connectivity of LIFC.
patterns obeyed the tripartite nature of language proces- Direct empirical support for this distribution of
sing (phonology, syntax, and semantics). These results labor between LIFC (Broca’s area) and temporal cortex
support the assumption of the functional division for was found in a study of Snijders et al. (2009). These
phonology, syntax, and semantics of the LIFC, includ- authors performed an fMRI study in which partici-
ing Broca’s area, and revealed a topographical func- pants read sentences and word sequences containing
tional organization in the left perisylvian language word-category (noun-verb) ambiguous words (e.g.,
network in which areas are most strongly connected “watch”) and the same materials with the unambigu-
according to information type (i.e., phonological, syntac- ous counterparts of the lexical-syntactic ambiguities.
tic, and semantic). The dorsal pathways might be more The ambiguous items were assumed to activate two
relevant for phonological and syntactic processing, independent syntactic frames, whereas the unambigu-
whereas the ventral pathways seem to be involved in ous counterparts result in the retrieval of only one syn-
connecting regions for semantic processing. tactic frame. Solely based on a computational model of
syntactic processing (Vosse & Kempen, 2000) and the
hypothesized contribution of temporal and frontal cor-
28.4 THE EMPIRICAL EVIDENCE FOR tex regions, it was predicted that the regions contribut-
THE MUC MODEL ing to the syntactic unification process should show
enhanced activation for sentences compared with
We have seen that there is a much more widespread words, and only within sentences should they display
connectivity profile in left perisylvian language cortex a larger signal for ambiguous than for unambiguous
than was assumed in the classical model. The MUC conditions. The posterior LIFC showed exactly this
model deviates from the classical model in the division predicted pattern, confirming the hypothesis that

D. LARGE-SCALE MODELS
 28.5 A GENERAL ACCOUNT OF THE ROLE OF LIFC IN LANGUAGE PROCESSING 343
LIFC, particularly BA 44 and BA 45, contributes to syn- that LIFC is a key node in the semantic unification net-
tactic unification. The left posterior MTG was activated work, unifying semantic information from different
more for ambiguous than unambiguous conditions, as modalities.
predicted for regions subserving the retrieval of From these findings it seems that syntactic and
lexical-syntactic information from memory. It thus semantic unification is realized in a dynamic interplay
seems that the LIFC is crucial for syntactic processing between LIFC as a multimodal unification site and
in conjunction with the left posterior MTG, a finding also knowledge-specific regions. Again, it is important
supported by patient studies with lesions in these very to stress that the interplay of these regions is crucial to
same regions (Caplan & Waters, 1996; Rodd, Longe, realize the functional component of unification.
Randall, & Tyler, 2010; Tyler et al., 2011). Presumably In other models, the anterior temporal lobe has been
these regions are connected via the dorsal pathways. argued to be relevant for combinatorial operations
In addition to syntactic unification, there is the need (Hickok & Poeppel, 2007; Rogalsky et al., Chapter 47
for semantic unification. One aspect of semantic unifi- of this volume). One possibility is that this is limited to
cation is filling the slots in an abstract event schema. conceptual combinations for which the mapping of
Semantic processing also recruits a left perisylvian grammatical roles (e.g., subject, object) onto thematic
network, albeit with a substantially weaker lateraliza- roles (e.g., agent, patient) is not required (Baron &
tion profile than syntactic processing. A series of fMRI Osherson, 2011). In the latter case, the contribution of
studies aimed to identify the semantic processing net- Broca’s region is presumably highly relevant.
work. These studies either compared sentences con-
taining semantic/pragmatic anomalies with their
correct counterparts (e.g., Friederici, Ruschemeyer, 28.5 A GENERAL ACCOUNT OF THE
Hahne, & Fiebach, 2003; Hagoort, Hald, Bastiaansen, & ROLE OF LIFC IN LANGUAGE
Petersson, 2004; Kiehl, Laurens, & Liddle, 2002; PROCESSING
Ruschemeyer, Zysset, & Friederici, 2006) or compared
sentences with and without semantic ambiguities So far, we have seen that LIFC plays a central role
(Davis et al., 2007; Hoenig & Scheef, 2005; Rodd, in syntactic and semantic unification processes, albeit
Davis, & Johnsrude, 2005). In the latter case, there are with different activation foci for these two types of
multiple word meanings for a given lexical item that unification. However, there is convincing evidence that
will induce competition and selection in relation to fill- LIFC also plays a role beneath the phrasal and sen-
ing a particular slot in the event schema. As with syn- tence level. It is found to contribute to decomposition
tactic unification, the availability of multiple and unification at the word level. Words are not pro-
candidates for a slot will therefore increase the unifica- cessed as unstructured, monolithic entities. Based on
tion load. In the case of the lexical-semantic ambigui- the morpho-phonological characteristics of a given
ties, there is no syntactic competition. Increased word, a process of lexical decomposition takes place in
processing is therefore attributable to unification of which stems and affixes are separated. For spoken
meaning instead of syntax. The most consistent finding words, the trigger for decomposition can be something
across studies on semantic unification is the activation as simple as the inflectional rhyme pattern, which is a
of the LIFC, particularly BA 47 and BA 45 (Hagoort & phonological pattern signaling the potential presence
Indefrey, 2014). of an affix (Bozic, Tyler, Ives, Randall, & Marslen-
A further indication for the contribution of LIFC in Wilson, 2010). Decomposing lexical input appears to
semantic unification comes from a few studies investi- be a ubiquitous and mandatory perceptual strategy;
gating semantic unification of multimodal information that is, decompositional processes are triggered not
with language. Using fMRI, Willems, Özyürek, and only for words with obvious parts (e.g., work-ed) but
Hagoort (2007) assessed the neural integration of also for semantically opaque words (e.g., bell-hop) and
semantic information from spoken words and from co- even nonwords with putative parts (e.g., blicket-s,
speech gestures into a preceding sentence context. blicket-ed). In a series of fMRI studies on the proces-
Spoken sentences were presented in which a critical sing of inflectional morphology, Bozic et al. (2010)
word was accompanied by a co-speech gesture. Either have found that LIFC, especially BA 45, subserves the
the word or the gesture could be semantically incon- process of morphological decomposition. Intracranial
gruous with respect to the previous sentence context. recordings in BA 45 from epileptic patients during pre-
Both an incongruous word as well as an incongruous surgical preparation indicate that the same brain area
gesture led to increased activation in LIFC (BA 45/47) is also involved in the generation of inflected forms
as compared with congruous words and gestures (for during language production (Sahin, Pinker, Cash,
a similar finding with pictures of objects, see Willems, Schomer, & Halgren, 2009; see also comments by
Őzyürek, & Hagoort, 2008). This supports the claim Hagoort & Levelt, 2009).

D. LARGE-SCALE MODELS
344 28. MUC (MEMORY, UNIFICATION, CONTROL): A MODEL ON THE NEUROBIOLOGY OF LANGUAGE BEYOND SINGLE WORD PROCESSING

The evidence for LIFC involvement at word- and unification by way of illustrating the principle of
sentence-level processing in both production and com- dynamic interaction.
prehension results in the question of how to account In posterior and inferior temporal and parietal
for its role more generally. This is still an open issue, (angular gyrus) regions, neuronal populations are acti-
but there is a possible answer. Notwithstanding the vated that represent lexical information associated
division of labor within LIFC, its overall contribution with the incoming word, including its semantic fea-
can be characterized in more general terms than hier- tures. From here, neural signals can follow two routes.
archical or even sentence-level processing. Instead, the The first exploits local connectivity within these poste-
LIFC is most likely involved in unification operations rior regions, resulting in a graded activation of neigh-
at the word and sentence level, in connection with boring neuronal populations, coding for related
temporal and parietal regions that are crucial for mem- lexical-semantic information. Such local spread of acti-
ory retrieval (Hagoort, 2005). Compositional and vation contributes to setting up a lexical-semantic con-
decompositional operations occur at multiple levels text in temporo-parietal cortex (Figure 28.5, green
and at multiple time slices in the language processing circle). The second route is based on long-distance con-
system, but also outside the language system. Any nections to LIFC, through direct white matter fibers,
time lexical and other building blocks enter into the resulting in the selective activation of populations of
process of utterance interpretation or construction, and frontal cortex neurons. These will respond with a self-
any time the input string requires decomposition (pre- sustaining firing pattern (see Durstewitz, Seamans, &
sumably through analysis-by-synthesis) to contact the Sejnowski, 2000 for a review). Efferent signals in this
right lexical representations, LIFC is recruited. case can only take the long-range route back. The most
This view is fully compatible with recent accounts parsimonious account here is that frontal neurons will
in linguistics that view both morphology and syntax to send efferent signals back to the same regions in
involve the retrieval of pieces of stored structure with temporo-parietal cortex from where afferent signals
variables (Jackendoff, personal communication, 2014). were received. This produces another spread of activa-
Hence, no principled distinction is claimed between tion to neighboring temporo-parietal regions, which
unification operations in syntax and morphology. implies that connections representing a given semantic
This account of LIFC’s contribution is more general
than is claimed in other models. For example, propo-
sals have been made that LIFC (Broca’s area) has a
more specialized role in language processing, has
more to do with linguistically motivated operations of
syntactic movement (Grodzinsky & Santi, 2008), and is
more involved in the processing of hierarchical struc-
tures (Friederici, Bahlmann, Heim, Schubotz, &
Anwander, 2006). However, such proposals are diffi-
cult to reconcile with the LIFC contributions to mor-
phological processes. Hence, the account specified here
seems to have more empirical support.

28.6 THE DYNAMIC INTERPLAY

BETWEEN MEMORY AND UNIFICATION

Although a connection is made between functional FIGURE 28.5 Processing cycle subserving semantic unification in
components of the cognitive architecture for language the left hemisphere language network. Inputs are conveyed from
sensory regions (here visual cortex) to the inferior, middle, and
and specific brain regions, this is an idealization of
superior temporal gyri (1), where lexical information is activated.
the real neurophysiological dynamics of the perisyl- Signals are hence relayed to the inferior frontal gyrus (2), where neu-
vian language network. Crucially, for language as for rons respond with a sustained firing pattern. Signals are then fed
most other cognitive functions, the functional contri- back into the same regions in temporal cortex from where they were
bution of any area or region has to be characterized in received (3). A recurrent network is thus set-up, which allows infor-
the context of the network as a whole, where speciali- mation to be maintained online, a context (green circle) to be formed
during subsequent processing cycles, and incoming words to be uni-
zation of any given node is only relative and realized fied within the context. At each processing cycle a balance is
in a dynamic interaction with the other nodes in the achieved by letting input-driven activity find attractor states, that is,
network (Mesulam, 1990, 1998). How this can be the maximum possible overlap with active populations in temporal
viewed is specified in more detail for semantic cortex.

D. LARGE-SCALE MODELS
 28.8 BEYOND THE CLASSICAL MODEL 345
context will be strengthened. During each word pro- tested the idea that pitch accent, which in Dutch is
cessing cycle, the memory (temporo-parietal) and uni- used to mark certain information as focus, recruits
fication (inferior frontal) components interact by attentional networks in the service of more extended
letting activation reverberate through the circuit in processing of the most relevant information. In our
Figure 28.5. Achieving the necessary outcomes for lan- study, the attentional network was first localized in an
guage comprehension may be more or less demanding, auditory nonverbal attention task. This task activated,
depending on how close the relation is between input as expected, bilateral superior and inferior parietal cor-
and context. tex. In the language task, participants were listening to
sentences with and sentences without semantic-
pragmatic anomalies. In half of the cases these anoma-
28.7 ATTENTIONAL CONTROL lies and their correct counterparts were marked as in
focus by a pitch accent; in the other half of the cases
The third component in the MUC model is referred they were not. The results showed an interaction in
to as Control. One form of control is attentional con- bilateral inferior parietal regions between prosody
trol. In classical models of sentence comprehension— (pitch accent) and congruence; for incongruent sen-
of either the syntactic-structure-driven variety (Frazier, tences, but not for congruent ones, there was larger
1987) or in a constraint-based framework (Tanenhaus, activation if the incongruent words carried a focus
Spivey-Knowlton, Eberhard, & Sedivy, 1995)—the marker (i.e., the pitch accent).
implicit assumption is usually that a full phrasal con- Overall, the activation overlap in the attention net-
figuration results and a complete interpretation of the works between the localizer task and the sentence
input string is achieved. However, often the listener processing task indicated that marking of information
interprets the input on the basis of bits and pieces that structure modulated a domain-general attention net-
are only partially analyzed. As a consequence, the lis- work. Pitch accent signaled the saliency of the
tener might overhear semantic information (the Moses focused words and thereby recruited attentional
illusion; Erickson & Mattson, 1981) or syntactic infor- resources for extended processing. This suggests that
mation (the Chomsky illusion; Wang, Bastiaansen, languages might have developed built-in linguistic
Yang, & Hagoort, 2012). In the question “How many devices (i.e., focus markers) that trigger the recruit-
animals of each kind did Moses take on the ark?”, peo- ment of attentional systems to safeguard against the
ple often answer “two,” without noticing that it was possibility that in a good-enough processing system
Noah who was in command of the ark, not Moses. It the most relevant information might go unnoticed.
was found that syntactic violations might not trigger a This provides one example of the interaction between
brain response if they are in a sentence constituent that a general demand/control system (Fedorenko,
provides no new information (Wang et al., 2012). Duncan, & Kanwisher, 2012) and the core components
Ferreira, Bailey, and Ferraro (2002) introduced the of the language network.
phrase “good-enough processing” to refer to the listen-
ers’ and readers’ interpretation strategies. In a good-
enough processing context, linguistic devices that 28.8 BEYOND THE CLASSICAL MODEL
highlight the most relevant parts of the input might
help the listener/reader in allocating processing I have outlined the contours of a neurobiological
resources optimally. This aspect of linguistic meaning model of language that is a substantial augmentation
is known as “information structure” (Buring, 2007; of the classical Wernicke-Lichtheim-Geschwind model,
Chafe, 1976; Halliday, 1967; Krifka, 2007). The informa- which was a model for single word processing mainly
tion structure of an utterance essentially focuses the based on lesion and patient data.
listener’s attention on the crucial (new) information in Three major additions are worth highlighting. First,
it. In languages such as English and Dutch, prosody the connectivity of the language cortex in left perisyl-
plays a crucial role in marking information structure. vian regions is much more extended than proposed in
For instance, in question answer pairs, the new or rel- the classical model and is certainly not restricted to the
evant information in the answer will typically be pitch arcuate fasciculus. Second, the distribution of labor
accented. After a question like “What did Mary buy at between the core regions in left perisylvian cortex is
the market?”, the answer might be “Mary bought not one in terms of production and comprehension.
VEGETABLES” (accented word in capitals). In this Shared circuitry has been established for core aspects of
case, the word “vegetables” is the focus constituent, language production and comprehension. Both recruit
which corresponds to the information provided for the temporal/parietal regions for retrieval of linguistic
Wh-element in the question. In a recent fMRI study information that is laid down in memory during acqui-
(Kristensen, Wang, Petersson, & Hagoort, 2013), we sition and LIFC for unification of building blocks into

D. LARGE-SCALE MODELS
346 28. MUC (MEMORY, UNIFICATION, CONTROL): A MODEL ON THE NEUROBIOLOGY OF LANGUAGE BEYOND SINGLE WORD PROCESSING

utterances or interpretations that are constructed online. Chafe, W. L. (1976). Givenness, contrastiveness, definiteness, sub-
Unification “enables words to cooperate to form new jects, topics and point of view. In C. N. Li (Ed.), Subject and topic.
New York, NY: Academic Press.
meanings” (Nowak, 2011, p. 179). Third, the operation Culicover, P. W., & Jackendoff, R. (2006). The simpler syntax
of language in its full glory requires a much more hypothesis. Trends in Cognitive Sciences, 10(9), 413 418.
extended network than what the classical model con- Davis, M. H., Coleman, M. R., Absalom, A. R., Rodd, J. M.,
tained, which was mainly based on evidence from single Johnsrude, I. S., Matta, B. F., et al. (2007). Dissociating speech per-
word processing. The basic principle of brain organiza- ception and comprehension at reduced levels of awareness.
Proceedings of the National Academy of Sciences of the United States of
tion for higher cognitive functions is that these are based America, 104(41), 16032 16037.
on the interaction between a number of neuronal circuits Davis, M. H., & Gaskell, M. G. (2009). A complementary systems
and brain regions that support the different contributing account of word learning: Neural and behavioural evidence.
functional components. These circuits are not necessarily Philosophical Transactions of the Royal Society of London. Series B,
specialized for language; nevertheless, they need to be 364, 3773 3800.
Durstewitz, D., Seamans, J. K., & Sejnowski, T. J. (2000). Dopamine-
recruited for the sake of successful language processing. mediated stabilization of delay-period activity in a network model
One example is the general attentional networks that of prefrontal cortex. Journal of Neurophysiology, 83(3), 1733 1750.
might be triggered into operation by specific linguistic Erickson, T. D., & Mattson, M. E. (1981). From words to meaning: A
devices to safeguard against missing out on the most rel- semantic illusion. Journal of Verbal Learning and Verbal Behavior, 20
evant (new, focused) information in the language input. (5), 540 551.
Fedorenko, E., Duncan, J., & Kanwisher, N. (2012). Language-
Another example, not further discussed here, is the selective and domain-general regions lie side by side within
Theory of Mind network that seems crucial for designing Broca’s area. Current Biology, 22(21), 2059 2062.
our utterances with knowledge of the listener in mind Fedorenko, E., & Thompson-Schill, S. L. (2014). Reworking the lan-
or, as a listener, to make the step from coded meaning to guage network. Trends in Cognitive Sciences, 18(3), 120 126.
speaker meaning (Bašnáková, Weber, Petersson, van Ferreira, F., Bailey, G. D. K., & Ferraro, V. (2002). Good-enough
representations in language comprehension. Current Directions in
Berkum, & Hagoort, 2013; Hagoort & Levinson, 2015). Psychological Science, 11(1), 11 15.
Frazier, L. (1987). Sentence processing: A tutorial review. In M.
Coltheart (Ed.), Attention and performance XII (pp. 559 585).
Acknowledgments London, UK: Erlbaum.
I am grateful to my colleagues in the NBL department and to Greg Friederici, A. D. (2009). Pathways to language: Fiber tracts in the
Hickok for their helpful comments. This chapter is an adapted ver- human brain. Trends in Cognitive Sciences, 13, 175 181.
sion of an earlier discussion of the MUC model (Hagoort, 2013). Friederici, A. D., Bahlmann, J., Heim, S., Schubotz, R. I., &
Anwander, A. (2006). The brain differentiates human and non-
human grammars: Functional localization and structural connec-
tivity. Proceedings of the National Academy of Sciences of the United
References States of America, 103(7), 2458 2463.
Anwander, A., Tittgemeyer, M., von Cramon, D. Y., Friederici, A. D., Friederici, A. D., Ruschemeyer, S. A., Hahne, A., & Fiebach, C. J.
& Knosche, T. R. (2007). Connectivity-based parcellation of (2003). The role of left inferior frontal and superior temporal cor-
Broca’s area. Cerebral Cortex, 17, 816 825. tex in sentence comprehension: Localizing syntactic and semantic
Baron, S. G., & Osherson, D. (2011). Evidence for conceptual combina- processes. Cerebral Cortex, 13(2), 170 177.
tion in the left anterior temporal lobe. Neuroimage, 55, 1847 1852. Grodzinsky, Y., & Santi, A. (2008). The battle for Broca’s region.
Bašnáková, J., Weber, K., Petersson, K. M., van Berkum, J., & Trends in Cognitive Sciences, 12(12), 474 480.
Hagoort, P. (2013). Beyond the language given: the neural corre- Hagoort, P. (2003). How the brain solves the binding problem for
lates of inferring speaker meaning. Cerebral Cortex, 24, 2572 2578. language: A neurocomputational model of syntactic processing.
Available from: http://dx.doi.org/10.1093/cercor/bht112. Neuroimage, 20, S18 S29.
Bozic, M., Tyler, L. K., Ives, D. T., Randall, B., & Marslen-Wilson, Hagoort, P. (2005). On Broca, brain, and binding: A new framework.
W. D. (2010). Bihemispheric foundations for human speech com- Trends in Cognitive Sciences, 9, 416 423.
prehension. Proceedings of the National Academy of Sciences of the Hagoort, P. (2009). Reflections on the neurobiology of syntax. In D.
United States of America, 107(40), 17439 17444. Bickerton, & E. Szathmáry (Eds.), Biological foundations and origin
Buring, D. (2007). Intonation, semantics and information structure. of syntax (pp. 279 299). Cambridge, MA: MIT Press.
In G. Ramchand, & C. Reiss (Eds.), The Oxford handbook of linguis- Hagoort, P. (2013). MUC (Memory, Unification, Control) and beyond.
tic interfaces. Oxford: Oxford University Press. Frontiers in Psychology, 4(Article 416). Available from: http://dx.
Caplan, D., & Waters, G. S. (1996). Syntactic processing in sentence doi.org/10.3389/fpsyg.2013.00416.
comprehension under dual-task conditions in aphasic patients. Hagoort, P., Hald, L., Bastiaansen, M., & Petersson, K. M. (2004).
Language and Cognitive Processes, 11, 525 551. Integration of word meaning and world knowledge in language
Caramazza, A., & Zurif, E. B. (1976). Dissociation of algorithmic and comprehension. Science, 304, 438 441.
heuristic processes in language comprehension: Evidence from Hagoort, P., & Indefrey, P. (2014). The neurobiology of language
aphasia. Brain and Language, 3(4), 572 582. beyond single words. Annual Review of Neuroscience, 37, 347 362.
Catani, M., Allin, M. P., Husain, M., Pugliese, L., Mesulam, M. M., Hagoort, P., & Levelt, W. J. M. (2009). The speaking brain. Science,
Murray, R. M., et al. (2007). Symmetries in human brain language 326(5951), 372 373.
pathways correlate with verbal recall. Proceedings of the National Hagoort, P., & Levinson, S. C. (2015). Neuropragmatics. In M. S.
Academy of Sciences of the United States of America, 104(43), Gazzaniga (Ed.), The cognitive neurosciences (5th ed.). Cambridge,
17163 17168. MA: MIT Press.

D. LARGE-SCALE MODELS
 REFERENCES 347
Halliday, M. A. K. (1967). Notes on transitivity and theme in English. Rodd, J. M., Davis, M. H., & Johnsrude, I. S. (2005). The neural
Part 2. Journal of Linguistics, 3, 177 274. mechanisms of speech comprehension: fMRI studies of semantic
Hickok, G., & Poeppel, D. (2004). Dorsal and ventral streams: A ambiguity. Cerebral Cortex, 15(8), 1261 1269.
framework for understanding aspects of the functional anatomy Rodd, J. M., Longe, O. A., Randall, B., & Tyler, L. K. (2010). Syntactic
of language. Cognition, 92, 67 99. and semantic processing of spoken sentences: An fMRI study of
Hickok, G., & Poeppel, D. (2007). The cortical organization of speech ambiguity. Paper presented at the 11th annual meeting Cognitive
processing. Nature Reviews Neuroscience, 8(5), 393 402. Neuroscience Society. San Francisco.
Hoenig, K., & Scheef, L. (2005). Mediotemporal contributions to Ruschemeyer, S. A., Zysset, S., & Friederici, A. D. (2006). Native and
semantic processing: fMRI evidence from ambiguity processing non-native reading of sentences: An fMRI experiment.
during semantic context verification. Hippocampus, 15(5), 597 609. Neuroimage, 31(1), 354 365.
Indefrey, P., & Cutler, A. (2004). Prelexical and lexical processing in Sahin, N. T., Pinker, S., Cash, S. S., Schomer, D., & Halgren, E.
listening. In M. S. Gazzaniga (Ed.), The cognitive neurosciences III (2009). Sequential processing of lexical, grammatical, and phono-
(3rd ed., pp. 759 774). Cambridge, MA: MIT Press. logical information within Broca’s area. Science, 326(5951),
Jackendoff, R. (2002). Foundations of language: Brain, meaning, grammar, 445 449.
evolution. Oxford, UK: Oxford University Press. Segaert, K., Menenti, L., Weber, K., Petersson, K. M., & Hagoort, P.
Joshi, A. K., & Schabes, Y. (1997). Treeadjoining grammars. In A. (2012). Shared syntax in language production and language com-
Salomma, & G. Rosenberg (Eds.), Handbook of formal languages and prehension—an FMRI study. Cerebral Cortex, 22(7), 1662 1670.
automata (Vol. 3, pp. 69 124). Heidelberg: Springer-Verlag. Snijders, T. M., Vosse, T., Kempen, G., van Berkum, J. J. A.,
Kelly, C., Uddin, L. Q., Shehzad, Z., Margulies, D. S., Castellanos, Petersson, K. M., & Hagoort, P. (2009). Retrieval and unification
F. X., Milham, M. P., et al. (2010). Broca’s region: Linking human of syntactic structure in sentence comprehension: An fMRI study
brain functional connectivity data and non-human primate tracing using word-category ambiguity. Cerebral Cortex, 19, 1493 1503.
anatomy studies. European Journal of Neuroscience, 32, 383 398. Sporns, O. (2011). Networks of the brain. Cambridge, MA: MIT Press.
Kiehl, K. A., Laurens, K. R., & Liddle, P. F. (2002). Reading Tanenhaus, M. K., Spivey-Knowlton, M. J., Eberhard, K. M., &
anomalous sentences: An event-related fMRI study of semantic Sedivy, J. C. (1995). Integration of visual and linguistic informa-
processing. Neuroimage, 17(2), 842 850. tion in spoken language comprehension. Science, 268(5217),
Krifka, M. (2007). Basic notions on information structure. In C. Féry, 1632 1634.
G. Fanselow, & M. Krifka (Eds.), Working papers of the SFB632, Tyler, L. K., Marslen-Wilson, W. D., Randall, B., Wright, P.,
interdisciplinary studies on information structure (ISIS) 6 (pp. 13 56). Devereux, B. J., Zhuang, J., et al. (2011). Left inferior frontal cortex
Potsdam: Universitätsverlag Potsdam. and syntax: Function, structure and behaviour in left-hemisphere
Kristensen, L. B., Wang, L., Petersson, K. M., & Hagoort, P. (2013). damaged patients. Brain, 134, 415 431.
The interface between language and attention: prosodic focus Vigneau, M., Beaucousin, V., Herve, P. Y., Jobard, G., Petit, L.,
marking recruits a general attention network in spoken language Crivello, F., et al. (2010). What is right-hemisphere contribution to
comprehension. Cerebral Cortex, 23(8), 1836 1848. Available from: phonological, lexico-semantic, and sentence processing? Insights
http://dx.doi.org/10.1093/cercor/bhs164. from a meta-analysis. Neuroimage, 54(1), 577 593.
Lau, E., Stroud, C., Plesch, S., & Phillips, C. (2006). The role of structural Vosse, T., & Kempen, G. A. M. (2000). Syntactic structure assembly
prediction in rapid syntactic analysis. Brain and Language, 98, 74 88. in human parsing: A computational model based on competitive
Lebel, C., & Beaulieu, C. (2009). Lateralization of the arcuate fascicu- inhibition and lexicalist grammar. Cognition, 75, 105 143.
lus from childhood to adulthood and its relation to cognitive abil- Wang, L., Bastiaansen, M. C. M., Yang, Y., & Hagoort, P. (2012).
ities in children. Human Brain Mapping, 30(11), 3563 3573. Information structure influences depth of syntactic processing:
McIntosh, A. R. (2008). Large-scale network dynamics in neurocogni- Event-related potential evidence for the Chomsky illusion. PLoS
tive function. In A. Fuchs, & V. K. Jirsa (Eds.), Coordination: One, 7(10), e47917.
Neural, behavioral and social dynamics (pp. 183 204). Springer: Willems, R. M., Özyürek, A., & Hagoort, P. (2007). When language
Berlin, Heidelberg. meets action: The neural integration of gesture and speech.
Menenti, L., Gierhan, S. M. E., Segaert, K., & Hagoort, P. (2011). Cerebral Cortex, 17, 2322 2333.
Shared language: Overlap and segregation of the neuronal infra- Willems, R. M., Őzyürek, A., & Hagoort, P. (2008). Seeing and hear-
structure for speaking and listening revealed by functional MRI. ing meaning: Event-related potential and functional magnetic res-
Psychological Science, 22, 1173 1182. onance imaging evidence of word versus picture integration into
Mesulam, M.-M. (1990). Large-scale neurocognitive networks and a sentence context. Journal of Cognitive Neuroscience, 20,
distributed processing for attention, language, and memory. 1235 1249.
Annals of Neurology, 28, 597 613. Wilson, S. M., Galantucci, S., Tartaglia, M. C., & Gorno-Tempini,
Mesulam, M.-M. (1998). Form sensation to cognition. Brain, 121, M. L. (2012). The neural basis of syntactic deficits in primary pro-
1013 1052. gressive aphasia. Brain and Language, 122(3), 190 198.
Nowak, M. (2011). Super cooperators: Beyond the survival of the fittest. Xiang, H., Fonteijn, H. M., Norris, D. G., & Hagoort, P. (2010).
Why cooperation, not competition is the key to life. Edinburgh- Topographical functional connectivity pattern in the perisylvian
London: Canongate. language networks. Cerebral Cortex, 20, 549 560.
Rilling, J. K., Glasser, M. F., Preuss, T. M., Ma, X., Zhao, T., Hu, X., et
al. (2008). The evolution of the arcuate fasciculus revealed with
comparative DTI. Nature Neuroscience, 11(4), 426 428.

D. LARGE-SCALE MODELS