PLEASE SCROLL DOWN FOR ARTICLE
This article was downloaded by: [University of Pennsylvania]On: 11 April 2011Access details: Access Details: [subscription number 932318042]Publisher Psychology PressInforma Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37-41 Mortimer Street, London W1T 3JH, UK
Language and Cognitive ProcessesPublication details, including instructions for authors and subscription information:http://www.informaworld.com/smpp/title~content=t713683153
Grounding the cognitive neuroscience of semantics in linguistic theoryLiina Pylkkänena; Jonathan Brennana; Douglas K. Bemisa
a Departments of Linguistics and Psychology, New York University, New York, NY, USA
First published on: 14 December 2010
To cite this Article Pylkkänen, Liina , Brennan, Jonathan and Bemis, Douglas K.(2010) 'Grounding the cognitiveneuroscience of semantics in linguistic theory', Language and Cognitive Processes,, First published on: 14 December 2010(iFirst)To link to this Article: DOI: 10.1080/01690965.2010.527490URL: http://dx.doi.org/10.1080/01690965.2010.527490
Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf
This article may be used for research, teaching and private study purposes. Any substantial orsystematic reproduction, re-distribution, re-selling, loan or sub-licensing, systematic supply ordistribution in any form to anyone is expressly forbidden.
The publisher does not give any warranty express or implied or make any representation that the contentswill be complete or accurate or up to date. The accuracy of any instructions, formulae and drug dosesshould be independently verified with primary sources. The publisher shall not be liable for any loss,actions, claims, proceedings, demand or costs or damages whatsoever or howsoever caused arising directlyor indirectly in connection with or arising out of the use of this material.
Grounding the cognitive neuroscience of semantics in
linguistic theory
Liina Pylkkanen, Jonathan Brennan, and Douglas K. Bemis
Departments of Linguistics and Psychology, New York University,
New York, NY, USA
The mission of cognitive neuroscience is to represent the interaction of cognitivescience and neuroscience: cognitive models of the mind guide a neuroscientificinvestigation of the brain bases of mental processes. In this endeavour, a cognitivemodel is crucial as without it, the cognitive neuroscientist does not know what tolook for in the brain, what the nature of the relevant representations might be, orhow the different components of a process might interact with each other. In thecognitive neuroscience of language, the interaction of theoretical models and brainresearch has, however, been far from ideal, especially when it comes to the study ofmeaning at the sentence level. Although theoretical semantics has a long history inlinguistics and thus offers detailed and comprehensive models of the nature ofsemantic representations, these theories have had minimal impact on the braininvestigation of semantic processing. In this article, we outline what a theoreticallygrounded cognitive neuroscience of semantics might look like and summarise ourown findings regarding the neural bases of semantic composition, the basiccombinatory operation that builds the complex meanings of natural language.
Keywords: Formal semantics; Cognitive neuroscience; Magnetoencephalography;
AMF.
If a visitor from Mars was told what the cognitive neuroscience of language is
and what linguistics is, they would presumably immediately assume that the
two disciplines must be tightly connected*after all, how could one study the
neural bases of language without an understanding of what language is? At
Correspondence should be addressed to Liina Pylkkanen, Departments of Linguistics and
Psychology, New York University, 10 Washington Place, New York, NY 10003, USA. E-mail:
The research discussed here was supported by the National Science Foundation Grant BCS-
0545186.
LANGUAGE AND COGNITIVE PROCESSES
0000, 00 (00), 1�21
# 2010 Psychology Press, an imprint of the Taylor & Francis Group, an Informa business
http://www.psypress.com/lcp DOI: 10.1080/01690965.2010.527490
Downloaded By: [University of Pennsylvania] At: 03:49 11 April 2011
the present time, however, the connection between the two fields remains
rather loose. Linguists have not done a terrific job in making their results
accessible to the general public, and perhaps neuroscientists have also not
reached out sufficiently to seek theoretical grounding for their research. Inour work, we seek to improve the situation for one particular subdomain of
language: combinatory semantics.
Of the various core processing levels of language, i.e., phonetics,
phonology, syntax, and semantics, the cognitive neuroscience of semantics
has been the most divorced from linguistics. For example, for phonology,
although there is no cognitive neuroscience for most questions that
phonologists are interested in, there is, for instance, a wide literature studying
the neural bases of phonological categories assuming exactly the types ofphoneme representations proposed by phonological theory (e.g., Dehaene-
Lambertz, 1997; Hickok & Poeppel, 2000; Naatanen et al., 1997). Similarly
for syntax, research probing into the function of Broca’s area, for example,
has assumed tree representations of the sort that one would also find in an
introductory syntax textbook (e.g., Caplan, Alpert, & Waters, 1999; Embick,
Marantz, Miyashita, O’Neil, & Sakai, 2000; Grodzinsky & Friederici, 2006;
Moro et al., 2001; Patel, 2003; Stromswold, Caplan, Alpert, & Rauch, 1996).
Not so for semantics. In fact, the word ‘‘semantics’’ tends to mean ratherdifferent things in cognitive neuroscience and in linguistics. In linguistics, the
semantics of an expression refers to the complex meaning that is computed by
combining the meanings of the individual lexical items within the expression.
The formal rules that are employed in this computation have been studied for
about 40 years, and many aspects of these rules are now well-understood. But
until recently, despite the clear necessity of understanding the neurobiology
of the combinatory functions that derive sentence meanings, we have had no
cognitive neuroscience on the brain bases of these rules. Instead, cognitiveneuroscience research with ‘‘semantic’’ in the title most often deals with the
representation of conceptual knowledge, investigating questions such as
whether living things and artifacts are distinctly represented in the brain (e.g.,
Mummery, Patterson, Hodges, & Price, 1998), or how different types of object
concepts are represented (Martin & Chao, 2001). The fact that this literature
is disconnected from linguistics is not devastating*the questions are rather
different from those that concern linguists. Although linguistic semantics
does include a subfield called ‘‘lexical semantics’’ that focuses on wordmeaning, even this subfield is mostly interested in word meaning that appears
to be composed of smaller parts*i.e., it has some type of complexity to it.
The representational differences between tools and animals are not a big
research question in linguistics and thus the brain research on conceptual
knowledge has less to gain from linguistic theories.
There is another line of brain research on ‘‘semantics’’, however, that
should and needs to connect with the types of theoretical models of meaning
2 PYLKKANEN, BRENNAN, BEMIS
Downloaded By: [University of Pennsylvania] At: 03:49 11 April 2011
representation offered by linguistics. This is the so-called ‘‘N400 literature’’,
which uses the N400 as an index of ‘‘semantic integration’’. Although this
functional interpretation is assumed by a sizeable research community (see
Lau, Phillips, & Poeppel, 2008 for a recent review), the N400 was not
discovered via a systematic search for a neural correlate of a theoretically
defined notion of ‘‘semantic integration’’. In fact, the behaviour and the
source localisation of the N400 are much more compatible with a lexical
access-based account (Lau et al., 2008), making the N400 an unlikely index
of semantic integrative processes.
In this article, we outline what we believe a theoretically grounded cognitive
neuroscience of semantics should look like. Our focus is on combinatory
semantics, i.e., the composition operations that serve to build complex
meanings from smaller parts. We take formal syntax and semantics of the
generative tradition (e.g., Heim & Kratzer, 1998) as the cognitive model that
guides this research and defines the operations whose neurobiology is to be
investigated. We assume, hopefully uncontroversially, that the right way to
uncover the neural bases of semantic composition is to systematically vary, or
otherwise track, this operation during language processing. In our own
research, we have aimed to do exactly this. We will first define exactly what we
mean by ‘‘semantic composition’’, then summarise our findings so far, discuss
some open questions, and finally, articulate our hopes for the future.
SEMANTIC COMPOSITION: DEFINING THE COREOPERATIONS
Theories of formal semantics are models of the possible meanings and
semantic combinatorics of natural language (e.g., Dowty, 1979; Heim &
Kratzer, 1998; Montague, 1970; Parsons, 1990; Steedman, 1996). They aim
to give a complete account of the representations and computations that
yield sentence meanings, including relatively straightforward rules that
combine predicates with arguments and adjuncts and extending to more
complicated phenomena such as the interpretation of tense, aspect, focus,
and so forth. The goal of these models is to understand and formalise
the nature of semantic knowledge both within languages and cross-
linguistically. Consequently, theories of formal semantics provide an
extremely rich and detailed hypothesis space both for psycho and neuro-
linguistic experimentation.
One challenge for rooting the cognitive neuroscience of semantics in
formal theories of meaning is deciding what theory to follow; unsurprisingly,
different models make different claims about semantic representations and
the mechanisms by which they are computed. Theoretical semantics is a
lively field with debate at every level of analysis, ranging from foundational
NEUROSCIENCE AND SEMANTIC THEORY 3
Downloaded By: [University of Pennsylvania] At: 03:49 11 April 2011
questions (such as what is the basic relationship between syntax and
semantics?) to issues relating to the formal details of the analysis of a
specific construction within a specific language. Such variance of opinion
may seem bewildering to a neuroscientist trying to figure out what aspects of
natural language meaning semanticists agree on, such that those phenomena
could safely be subjected to a brain investigation.
Unfortunately, there is no safe route: like any other branch of cognitive
science, semantic theory is ever evolving and virtually every aspect of it has
been or at least can be questioned. However, it is still possible to identify
certain basic operations as a starting point, while keeping in mind the
possibly controversial aspects of the formal treatment. In what follows, we
describe two operations that form the core of the combinatory engine in
most semantic theories (for a fuller exposition, see Pylkkanen & McElree,
2006). Our notation follows Heim and Kratzer (1998), but again, for our
present purposes the details of the rules are not important; rather, what is
crucial is that this type of theory distinguishes between two different
modes of composition, one that serves to fill in argument positions of
predicates and one that modifies the predicate without impacting its
arguments. Although mainstream, this distinction is neither a formal
necessity nor necessarily empirically correct (e.g., Pietroski, 2002, 2004,
2006), but given its central role in most semantic theories, it is one of the
most basic distinctions that can be subjected to a brain investigation. Since
one reason for the disconnection between cognitive neuroscience and formal
semantics has likely been the relative impenetrability of semantic theories for
the nonlinguist, in the following we aim for a maximally informal and
accessible description of the basic ideas.
Function application
The driving force behind mainstream theories of semantic composition is
the idea that the meanings of most words are in some sense incomplete, or
‘‘unsaturated’’, unless they are combined with other words with suitable
meanings (Frege, 1892). For an intuitive example, the semantics of action
verbs is thought to have ‘‘placeholders’’, or variables, that stand for the
participants of the actions described by the verbs. In order for the meanings
of these verbs to be complete, or saturated, the verb needs to combine with
noun phrases that describe those participants. More formally, these verbs are
treated as functions that take individual-denoting noun phrases as their
arguments. This idea is expressed in lambda calculus for the transitive verb
destroy in (1) below. The arguments of the function are prefixed with
lambdas, and the value, or the output of the function, follows the lambda
terms:
4 PYLKKANEN, BRENNAN, BEMIS
Downloaded By: [University of Pennsylvania] At: 03:49 11 April 2011
(1) destroy: lx.ly. destroy(y, x)
In order to combine the verb destroy with its arguments, we apply a
rule called function application, which essentially replaces the variables in the
denotation of the predicate and erases the lambda prefixes. Proper names
are the most straightforward individual-denoting noun phrases and thus the
representation of the sentence John destroyed Bill would involve roughly the
following representation (here we ignore everything that does not pertain to
the argument saturation of the verbal predicate).
(2)
In addition to simple examples such as the one above, function application is
the mode of composition for any predicate and its argument. For example,
prepositions combine with noun phrases via function application [in [the dark]]
as do verbs with clausal complements [knew [that John destroyed Bill]] and
verbal ones [John tried [to destroy Bill]]. Thus function application accounts
for a vast amount of semantic composition in natural language and every
theory of semantics requires a rule that accomplishes its task. Consequently, it
should be regarded as one of the core operations that any theory of the brain
bases of language should address.
Predicate modification
The gist of function application is that it satisfies the semantic requirements
of a predicate and makes it more complete. But most expressions also have
parts that are not formally required, i.e., they could be dropped and the
expression would still be grammatical. For instance, although we could
describe The New York Times simply as a newspaper, we could also describe
it as a major national newspaper. There are no grammatical reasons for why
we would ever have to add these adjectives*unlike with destroy above, which
simply could not function grammatically without, say, its object (*John
destroyed). The adjectives simply provide additional, potentially useful
information about the noun. The majority of these types of optional
elements are added into expressions via a rule called predicate modification.
NEUROSCIENCE AND SEMANTIC THEORY 5
Downloaded By: [University of Pennsylvania] At: 03:49 11 April 2011
Predicate modification builds complex properties from simpler ones. In
formal semantics, adjectives and nouns are both treated as describing
properties of individuals, e.g., newspaper denotes the property of being a
newspaper, and major the property of being major. Predicate modification
simply concatenates these properties, such that major national newspaper
ends up meaning ‘‘the property of being major, and national, and a
newspaper’’, as shown in lambda notation in (3).
(3)
In addition to building complex properties of individuals, predicate modifica-
tion can also combine any two predicates that describe properties of the same
type of thing. For example, in event-based semantic frameworks (e.g.,
Davidson, 1967; Kratzer, 1996; Parsons, 1990), both verbs and adverbs are
thought of as describing properties of events. Thus the mode of composition
between, say, run, and quickly would be a version of predicate modification,
such that the resulting verb phrase would describe events that are both quick
and events of running.
In summary, while function application is the default mode for adding
arguments into a structure, predicate modification is the most basic way to
add optional elements. Although natural language semantics as a whole
requires a much more elaborate system than just these two rules, function
application and predicate modification constitute the nuts and bolts of
semantic composition in most theories, and thus offer clearly defined
computations whose neural bases can be characterised within the cognitive
neuroscience of language. It is of course an empirical question whether
function application and predicate modification dissociate in brain activity.
If they do not, this would be a type of null result and not necessarily a
motivation for linguists to revise their theories, but a lack of dissociation
would obviously call into question the assumption that the operations are
distinct. In this type of situation, the brain data would be more compatible
with a theory where semantic composition is achieved via a unified operation
across expression types, as proposed, for example, by Pietroski (2002, 2004,
2006). In the long run, this is the type of interaction between brain
experimentation and linguistic theorising that we advocate.
6 PYLKKANEN, BRENNAN, BEMIS
Downloaded By: [University of Pennsylvania] At: 03:49 11 April 2011
SEMANTIC REPRESENTATIONS VS. THEIR PLAUSIBILITYGIVEN WORLD KNOWLEDGE
Theories of formal semantics have a crisp definition of semantic well-
formedness: an expression is semantically well-formed if the rules of
composition yield a meaning for it. For example, in the default case, there
is no way for a verb such as arrive, i.e., a function that takes a single
individual-denoting argument, to combine with two such arguments, given
that this predicate only has one argument slot for function application
to saturate. Thus an expression such as I arrived Helsinki is considered
uninterpretable under this type of theory; to make it interpretable, another
predicate would need to be added, such as the preposition in, which can then
take Helsinki as its argument.This definition of well-formedness is very different from layman’s
intuition of what ‘‘makes sense’’, i.e., an expression can be semantically
well-formed in a formal sense even if it means something crazy. For example,
I just ate a cloud is a nonsensical sentence in that the situation it describes
could never happen (unless the world was significantly altered). However,
every native speaker of English should have the intuition that they know
what this sentence means and, in fact, this meaning is easily computed by the
two rules introduced above. In terms of formal semantics, the oddness of the
sentence is not revealed until one compares its semantic representation
(computed by the rules) to one’s knowledge of the world, which dictates that
clouds are not edible entities. But crucially, the sentence is not semantically
ill-formed, but rather just ill-fitting to our world knowledge.
This conception of semantic well-formedness is dramatically different
from what is generally assumed in neuroimaging and especially in event-
related potentials (ERP) research on semantics. Such brain research has been
overwhelmingly dominated by studies that vary ‘‘semantic well-formedness’’
or ‘‘congruity’’, but in the majority of these experiments, the semantically ill-
formed stimuli are not ill-formed in the above described, formal sense; rather,
in terms of linguistic theory, they violate world knowledge instead. The
original ERP study on such violations, by Kutas and Hillyard (1980), used
sentences such as he spread the warm bread with socks to violate what they
called ‘‘semantic appropriateness’’. These violations elicited a negative-going
ERP peaking around 400 ms, the N400, a finding that has since been
replicated in hundreds of studies. Since the N400 is elicited by semantically
surprising elements, it is commonly thought of as a semantics-related ERP.
However, this notion of ‘‘semantics’’ is not theoretically grounded, i.e., we
are not aware of any theory of linguistic representation according to which
typical N400 sentences would count as semantically ill-formed (and even if
there is such as a theory, the N400 literature does not reference it). For
instance, in the above example he spread the warm bread with socks, the final
NEUROSCIENCE AND SEMANTIC THEORY 7
Downloaded By: [University of Pennsylvania] At: 03:49 11 April 2011
word socks composes without a problem with the preposition with, to yield a
meaning in which socks are used as a spread on bread. Now socks are of
course not edible or spreadable, but in terms of linguistic theory, this is a fact
about the world, not a fact about language. Given that there is no crisp
definition of what ‘‘semantics’’ means in the N400 literature, considering
this component a semantics-related ERP is not conducive to building a
theoretically grounded model of language in the brain.
It is an empirical question whether N400-type effects could also be
obtained for expressions that are semantically ill-formed in the linguistic
sense. There are a few studies that have run relevant manipulations, primary
focusing on violations of negative polarity licensing (Drenhaus, beim Graben,
Saddy, & Frisch, 2006; Saddy, Drenhaus, & Frisch, 2004; Steinhauer, Drury,
Portner, Walenski, & Ullman, 2010; Xiang, Dillon, & Phillips, 2009). Negative
polarity items (NPIs) are expressions such as ever or any that must occur in the
scope of negation1 (e.g., No tourist ever visited the old harbour vs. *One tourist
ever visited the old harbour). Violations of this licensing condition count as true
semantic violations in terms of linguistic theory, in that if an NPI is not
licensed by a suitable operator, no semantic representation can be built for the
sentence. Negative polarity violations have indeed been reported to elicit N400
effects (Saddy et al., 2004), but also P600 effects (Drenhaus et al., 2006; Xiang
et al., 2009) and late left anterior negativities (Steinhauer et al., 2010). Thus
the ERP profile of these violations has been quite variable and does not easily
lend itself to firm conclusions. Most relevantly for our present purposes,
however, negative polarity violations have exhibited rather different ERP
profiles from world knowledge violations when the two are examined within
the same study (Steinhauer et al., 2010).
In summary, most extant research on the cognitive neuroscience of
sentence level meaning has operated with only an intuitive notion of
‘‘semantics’’, not grounded in any particular theoretical model. It should,
however, be uncontroversial that technical terminology in any field needs to
be clearly defined; otherwise it is unclear what the research is even about.
Given the vast amount of work devoted to carefully characterising crisp,
formal semantic theories, disconnecting the cognitive neuroscience of
semantics from these models is a missed opportunity. Only by grounding
the neurobiology of meaning in theoretical models of meaning can we move
forward to a ‘‘predictive mode’’ of research, where the properties of a
cognitive model guide the experimentation and make predictions about what
phenomena should group together. Without such grounding, one is left with
complex data-sets and a layman’s intuition about what it all may mean.
1 Or in the scope of some other operator that supports decreasing inferences (Ladusaw, 1979,
1980, 1983).
8 PYLKKANEN, BRENNAN, BEMIS
Downloaded By: [University of Pennsylvania] At: 03:49 11 April 2011
THE CHALLENGE OF COMPOSITIONALITY
As soon as one couches the brain investigation of semantics in linguistic
theory, a methodological dilemma, however, arises: In most cases, the
semantic complexity of an expression tightly correlates with its syntactic
complexity, and thus it is not obvious how one could ever vary semantic
composition without syntactic confounds. Even worse, linguistic theories
differ in the extent to which syntax and semantics are coupled. In some
theories, every step of the semantic derivation corresponds to a step in the
syntactic derivation (e.g., Heim & Kratzer, 1998; Montague, 1970). Other
theories assume a weaker coupling between syntax and semantics, allowing
certain operations to only apply in the semantics without a corresponding
syntactic step (Barker, 2002; Hendriks, 1988; Jacobson, 1999; Partee & Rooth,
1983). And finally, in Jackendoff ’s parallel model, syntactic and semantic
representations are built by completely independent mechanisms, creating an
architecture that forces no correlation between the number of syntactic and
semantic steps (Jackendoff, 2002).2 The disagreement between these theories
has to do with the degree of compositionality in natural language, i.e., the
extent to which the meanings of expressions are straightforwardly determined
by the meanings of their parts and the syntactic combinatorics.
With this type of uncertainty about the fundamental relationship between
syntax and semantics, how could a cognitive neuroscientist use these theories
to make headway in understanding the brain bases of semantic composition?
In our own research, we have adopted what might be called a ‘‘bite the
bullet’’ approach. First, we have to accept that there are no expressions that
uncontroversially involve semantic computations that do not correspond to
any part of the syntax: even if some meaning component of an expression
does not appear to map onto the syntax, one can always postulate a
phonologically null syntactic element that carries that piece of meaning. The
question though is, whether such an analysis makes the right empirical
predictions. As discussed in Pylkkanen (2008), there is great asymmetry in
the degree to which this type of solution works for different cases; it
sometimes makes the right predictions, but often not. As a starting point in
our research, we have aimed to study exactly the expressions that are most
likely to involve syntactically unexpressed semantic operations, even if this
dissociation cannot be definitively proven. As discussed in the next section,
our subsequent research has shown that these types of cases pattern
differently from cases where syntax and semantics are uncontroversially
coupled, lending support to the notion that the initial set of test cases did, in
2 Although such a correlation can and must be obviously be built in, given the descriptive
generalisation that syntactic and semantic operations do, by and large, stand in a one-to-one
correspondence.
NEUROSCIENCE AND SEMANTIC THEORY 9
Downloaded By: [University of Pennsylvania] At: 03:49 11 April 2011
fact, only vary semantics. Thus compositional coupling of syntax and
semantics makes the project of ‘‘neurosemantics’’ a difficult one to get off the
ground, but once some initial progress is made, we believe it may be possible
for brain data to actually contribute to debates about the relationshipbetween syntax and semantics.
THE ANTERIOR MIDLINE FIELD (AMF) AS A CORRELATE OFSEMANTIC COMPOSITION
Varying semantic composition: semantic mismatch
Although much of natural language appears strongly compositional, there
are classes of expressions whose meanings appear richer than their syntax.Perhaps the best studied example is so-called complement coercion, illustrated
in (4) below.
(4) a. The professor began the article.
b. The boy finished the pizza.
c. The trekkers survived the mountain.
When native speakers are queried about the meanings of these sentences, theytypically report for (4a) that the professor began reading the article, (4b) that
the boy finished eating the pizza, and (4c) that the trekkers survived climbing
the mountain. Reading, eating, or climbing do not, however, figure in the
lexical material of these expressions. Thus where do these implicit activity
senses come from? They are typically thought of as arising from a certain
semantic mismatch between the verbs and their direct objects. Semantically,
each of the verbs in (4) selects for a direct object that describes some type of
event or activity: one generally begins, finishes, or survives doing something.However, none of the direct objects in (4) describe events, rather, they all
denote entities. This kind of situation is formally considered a semantic ‘‘type
mismatch’’ and in the default case, type mismatch leads to ungrammaticality.
However, the sentences in (4) are all grammatical and thus the type mismatch
must be somehow resolved. Descriptively, the resolution appears to involve
the semantic insertion of an implicit activity that can mediate between
the verb and the otherwise ill-fitting object NP. Formally, the complement of
the verb (the object NP) is thought to ‘‘coerce’’ into an event meaning(Jackendoff, 1997; Pustejovsky, 1995), such that semantic composition can
succeed. This type of analysis treats coercion as a purely semantic meaning-
adding operation, with no consequences for the syntax.3
3 For an extensive discussion on the possible empirical arguments for this, see Pylkkanen
(2008).
10 PYLKKANEN, BRENNAN, BEMIS
Downloaded By: [University of Pennsylvania] At: 03:49 11 April 2011
Coercion does, however, have consequences for online processing:
numerous studies have shown that coercion expressions take longer to process
than control expressions involving no coercion (for a review, see Pylkkanen &
McElree, 2006). This psycholinguistic evidence provides empirical support forthe hypothesis that coercion expressions do, in fact, involve some type of extra
computation, absent in more transparently compositional expressions. Given
that the processing of complement coercion had already been well-studied
behaviourally, we used it as the initial test case in our attempt to identify a
neural correlate of purely semantic combinatoric processing.
To measure brain activity, our research uses magnetoencephalography
(MEG), which offers the best combination of both spatial and temporal
resolution of existing cognitive neuroscience techniques. MEG measures themagnetic fields generated by neuronal currents. The primary difference
between MEG and EEG is that in MEG, the locations of the current
generators can be estimated reasonably accurately given that magnetic fields
pass through the different structures of the brain undistorted, contrary to
electric potentials. Typically, the current sources of MEG recordings as
modelled either as focal sources (so-called single dipoles) or as distributed
patches of activation on the cortex (typically minimum norm estimates)
(Hansen, Kringelbach, & Salmelin, 2010). The intensities and latencies ofthese current sources then function as the primary dependant measures of
most MEG studies, although an EEG-style analysis of sensor data without
source localisation is also always an option.
When we contrasted coercion expressions (the journalist began the article)
with noncoercive control sentences (the journalist wrote the article) during an
MEG recording, we observed increased activity for coercion in a prefrontal
midline MEG field, dubbed the anterior midline field (AMF) (Figure 1).
Source localisation indicated the ventromedial prefrontal cortex (vmPFC)as the generating brain region of this magnetic field. No such effect was
observed for implausible control sentences, suggesting different brain bases
for the computation of coercion and the detection of poor real-world fit.
These complement coercion findings established a starting point for our
research on the neural bases of semantic composition. Our subsequent studies
have then aimed to narrow down the possible functional interpretations of this
activity. Figure 1 summarises all of our AMF results so far. First, we examined
whether the AMF effect generalises to other coercion constructions, andfound that it is also observed for a different variant of complement coercion
(Pylkkanen, Martin, McElree, & Smart, 2009) as well as for two different types
of aspectual coercion (Brennan & Pylkkanen, 2008, 2010). These findings
ruled out the hypothesis that the AMF reflects processing specific to
complement coercion.
We have also examined the AMF in a violation paradigm (Pylkkanen,
Oliveri, & Smart, 2009), to better connect our findings to ERP research
NEUROSCIENCE AND SEMANTIC THEORY 11
Downloaded By: [University of Pennsylvania] At: 03:49 11 April 2011
which has been dominated by this type of design. In this study, we again used
semantic mismatch, but of a sort that cannot be resolved by a productive
rule, but rather results in semantic ill-formedness (in the linguistic sense, i.e.,
no well-formed representation can be built). These semantic violations were
contrasted both with violations of world knowledge (similar to the ‘‘semantic
violations’’ of the N400 literature) and with well-formed control expressions,
Figure 1. Summary of all our AMF results to date, including the typical MEG response to a
visual word presentation in a sentential context (top panel). The upper left corner depicts the
AMF field pattern and a typical localization. In the stimulus examples, the critical word is
underlined.
12 PYLKKANEN, BRENNAN, BEMIS
Downloaded By: [University of Pennsylvania] At: 03:49 11 April 2011
in order to assess in an ERP-style violation paradigm whether semantic
violations but not world knowledge violations would affect the AMF, as
would be predicted if this activity reflected the composition of semantic
representations but not the evaluation of their real-world fit. Our results
indeed patterned according to this prediction: semantic violations elicited
increased AMF amplitudes, while world knowledge violations generated a
different type of effect.
Thus our initial set of experiments employed resolvable and unresolvable
semantic mismatch in order to vary semantic composition while keeping
syntax maximally constant. These studies yielded highly consistent
results, implicating the AMF MEG component as potentially relevant for
the construction of complex meaning. This effect demonstrated task-
independence (Pylkkanen, Martin, et al., 2009) and was insensitive to world
knowledge (Pylkkanen, Oliveri, et al., 2009). But of course these results do not
yet show that we have isolated activity reflecting semantic composition in some
general sense, as opposed to activity that plays a role in mismatch resolution
but does not participate in more ordinary composition. Furthermore, the
psychological nature of coercion operations is vastly underdetermined by
traditional linguistic data (e.g., judgements of grammaticality and truth value);
in some sense ‘‘coercion’’ and ‘‘type-shifting’’ are descriptive labels for
computations that matter for interpretation but do not appear to have a
syntactic nature. In other words, although coercion rules are traditionally
thought of as operating within the compositional semantic system, it is difficult
to demonstrate this empirically, i.e., for example, to rule out the possibility that
these meaning shifts might be essentially pragmatic in nature. Having
discovered that coercion affects a particular brain response, i.e., the AMF, it
becomes possible to investigate whether the same response would also be
affected by simpler manipulations of semantic composition, involving no
mismatch resolution. If ordinary ‘‘run-of-the-mill’’ composition was to also
modulate the AMF, this would show that the role of the AMF is not limited
to mismatch resolution and would also offer empirical support for the
hypothesis that coercion operations involve semantic computations similar
to those involved in transparently compositional expressions. The research
summarised below aimed to assess this by studying very simple cases of
composition, involving only the intersective combination of a noun and an
adjective.
Simple composition: bringing in syntax and the left anteriortemporal lobe (ATL)
Our study on simple composition constitutes, to our knowledge, the first
neurolinguistic investigation directly targeting one of the core semantic
operations outlined in Section ‘‘Semantic composition: defining the core
NEUROSCIENCE AND SEMANTIC THEORY 13
Downloaded By: [University of Pennsylvania] At: 03:49 11 April 2011
operations’’ above. Specifically, we aimed to measure the MEG activity
associated with predicate modification (Bemis & Pylkkanen, 2010). Contrary
to most extant brain research on syntax and semantics, which has generally
employed full sentences involving complex structures such as centreembedding or semantic anomalies (for reviews, see Kaan & Swaab, 2002;
Lau et al., 2008), our study employed minimal phrases invoking exactly one
step of semantic composition. The critical stimulus was an object-denoting
noun that was either preceded by a semantically composable colour adjective
(red boat) or by consonant string activating no lexical meaning (xkq boat).
Subjects viewed these phrases (and other control stimuli) and then decided
whether or not a subsequent picture matched the verbal stimulus. If the role
of the AMF is limited to mismatch resolution, it clearly should not beaffected by this maximally simple manipulation. In contrast, if the AMF
reflects aspects of semantic composition quite generally, nouns preceded by
adjectives should elicit increased AMF activity. The latter prediction was
robustly confirmed: a clear AMF amplitude increase was observed for the
adjective�noun combinations relative to the isolated nouns, ruling out the
hypothesis that the AMF is purely mismatch related.
But predicate modification is of course not the only combinatory
operation varied in this manipulation: each adjective�noun pair also madeup a syntactic phrase. Thus the above contrast should also elicit effects
related to the syntactic side of the composition, if this is indeed something
separable from the semantic combinatorial effects. We did, in fact, also
observe a second effect for the complex phrases, and in a location familiar
from a series of prior studies. This effect localised in the left anterior
temporal lobe (ATL), which has been shown by several imaging studies as
eliciting increased activation for sentences as opposed to unstructured lists of
words (Friederici, Meyer, & von Cramon, 2000; Humphries, Binder, Medler,& Liebenthal, 2006; Mazoyer et al., 1993; Rogalsky & Hickok, 2008; Stowe
et al., 1998; Vandenberghe, Nobre, & Price, 2002;). This body of work has
hypothesised that the left ATL is the locus of basic syntactic composition,
with Broca’s region only participating in more complex operations. This
interpretation is further corroborated by imaging results of our own,
showing that while subjects listen to a narrative, activity in the left ATL
correlates with the number of syntactic constituents completed by each word
(Brennan et al., 2010). Thus the combination of the results reviewed so farsuggests the following working hypothesis: the AMF generator, i.e., the
vmPFC, and the left ATL are the primary loci of basic combinatorial
processing, with the vmPFC computing semantic and the left ATL syntactic
structure.
The hypothesis just articulated raises a puzzle, though, regarding the
relationship between our results on the vmPFC and the just mentioned
imaging studies contrasting sentences vs. word lists. Clearly, the sentence vs.
14 PYLKKANEN, BRENNAN, BEMIS
Downloaded By: [University of Pennsylvania] At: 03:49 11 April 2011
word list contrast varies not only syntactic composition but also semantic
composition, yet none of the studies using this contrast have reported effects in
the vmPFC. This is obviously incompatible with our semantic composition
hypothesis regarding this region. But the sentence vs. word list studies have alsoused a different technique from our research, measuring slowly arising
hemodynamic activity as opposed to electromagnetic activity, which can be
measured at a millisecond temporal resolution, matching the speed of language
processing. To assess whether a vmPFC effect for sentences over word lists
would be observed in MEG, we conducted an MEG version of the traditional
imaging design (Brennan & Pylkkanen, 2010). Our results indeed revealed an
amplitude increase in the vmPFC for sentences, conforming to the composi-
tion hypothesis. An effect was also seen in the left ATL, replicating the priorimaging results. While questions remain about why hemodynamic methods
and MEG should yield different results here, the picture emerging from our
MEG findings is encouraging and begins to have the shape of an elementary
theory about the neural bases of syntactic and semantic composition,
grounded in linguistic theory.
Challenges and open questions
Relation to hemodynamic and neuropsychological data
Our MEG research so far suggests a prominent role for the generating
region of the AMF in language processing. However, if the role of this
midline prefrontal region is as basic as semantic composition, how come this
area has not already figured in neuroimaging studies on sentence processing?
As just discussed, it is possible that MEG may be better suited for capturing
this activity than hemodynamic methods, given our findings on the sentence
vs. word list paradigm which yields a vmPFC effect in MEG but not in fMRIor PET (granted that we have not yet conducted a study with the same
materials and subjects across the techniques). One possible reason for this
difference lies in the better time resolution of MEG: a short-lived time-
locked effect lasting less than 100 ms should naturally be difficult to capture
with techniques whose temporal resolution is on the scale of several seconds.
Furthermore, fMRI in particular is limited when it comes to measuring
signal from the vmPFC and the orbitofrontal cortex in general. This is due to
the proximity of these regions to the sinuses, which leads to so-calledsusceptibility artifacts, resulting in significant signal loss in orbitofrontal
regions (Ojemann et al., 1997). Despite these limitations, however, several
hemodynamic studies have shown interpretation-related effects in the
vmPFC, both in PET (e.g., Maguire, Frith, & Morris, 1999; Nathaniel-
James & Frith, 2002) and in fMRI (Nieuwland, Petersson, & Van Berkum,
2007). Thus it is not the case that our findings are entirely without precedent
in the neuroimaging literature.
NEUROSCIENCE AND SEMANTIC THEORY 15
Downloaded By: [University of Pennsylvania] At: 03:49 11 April 2011
One type of evidence that does not suffer from the artifact issues described
above pertains to neuropsychological data. Do patients with ventromedial
prefrontal damage suffer severe language processing deficits, of the sort one
would expect of a person who has lost their ability to construct complexlinguistic meanings? The most simple-minded prediction of the semantic
composition hypothesis might be that such patients should not be able to
understand anything or be able to produce any coherent sentences. This
prediction is certainly not born out: language problems are not the
predominant issue for this patient population. Instead, their most salient
impairments have to do with appropriateness of social behaviour, decision-
making, and emotion regulation (Anderson, Barrash, Bechara, & Tranel,
2006; Barrash, Tranel, & Anderson, 2000; Berlin, Rolls, & Kischka, 2004;Burgess & Wood, 1990; Grafman et al., 1996; Koenigs & Tranel, 2007).
Perhaps because language problems do not present prominently, language
skills are typically not even reported, or they may be informally commented
on as ‘‘normal’’ (e.g., Anderson et al., 2006). There is though one body of
research on vmPFC patients that has focused on language processing,
specifically on sarcarm and irony. In this work, vmPFC damage has been
found to lead to deficits in comprehending sarcasm, a finding that has
been related to the role of the vmPFC in theory-of-mind processing(Shamay-Tsoory, Tibi-Elhanany, & Aharon-Peretz, 2006; Shamay-Tsoory,
Tomer, & Aharon-Peretz, 2005). This finding is obviously related to semantic
processing, but not at the very basic level that our most recent simple
composition results suggest (Section ‘‘Simple composition: bringing in
syntax and the left anterior temporal lobe (ATL)’’).
What then, to make of these relatively sparse neurolinguistic data on
vmPFC patients in light of the semantic composition hypothesis of the
AMF? At least two considerations are important to note here, onemethodological and the other theoretical. The methodological consideration
is that although MEG systematically localises the AMF in the vmPFC (with
some variation on the anterior�posterior axis), this localisation is a source
model and not necessarily the absolute truth. For example, so far most of our
source localisations have used a smoothed average cortex (provided by
BESA, the source localisation software), which does not include the medial
wall. In this kind of source model, activity in the anterior cingulate, for
example, might be projected onto vmPFC. Thus it is not yet obvious thatvmPFC patients are necessarily the right clinical population to consider.
Clearly, the localisation of the AMF will need to be further assessed with
hemodynamic methods, although as just reviewed, this approach presents
challenges of its own.
The theoretical consideration has to do with the precise predictions of
the semantic composition hypothesis. What should a person’s linguistic
performance look like if they have lost their ability of semantic composition
16 PYLKKANEN, BRENNAN, BEMIS
Downloaded By: [University of Pennsylvania] At: 03:49 11 April 2011
but have intact syntax and lexical-semantic processing? The answer is far from
obvious. Such a person would have a combinatory system, i.e., the syntax, and
they would understand the meanings of words. Thus it is possible that they
might be able to reasonably work out what sentences mean, only failing insubtler situations. Thus perhaps the only clear prediction of the semantic
composition hypothesis is that a person without the relevant region should
show some semantic deficits. Whether vmPFC patients fit this profile is
currently unclear; assessing this would require sophisticated linguistic testing
perhaps focusing on semantic phenomena that are not transparently reflected
in the syntax.
Domain generality
One of the most fundamental questions regarding the neural architecture of
language is the extent to which the mechanisms employed by language are
specific to language or also used in other domains. Localisation wise, ourfindings on the possible brain bases of semantic composition have been quite
surprising: the midline prefrontal regions that are modelled as the AMF source
are not generally thought of as ‘‘language regions’’. Instead, these areas,
and the vmPFC specifically, are prominently associated with various types
of nonlinguistic higher cognition, such as emotion (Bechara, Damasio, &
Damasio, 2000; Damasio, 1994), decision-making (Wallis, 2007), representa-
tion of reward value (Schoenbaum, Roesch, & Stalnaker, 2006), and social
cognition, including theory-of-mind (Amodio & Frith, 2006; Baron-Cohen &Ring, 1994; Baron-Cohen et al., 1994; Gallagher & Frith, 2003; Krueger,
Barbey, & Grafman, 2009; Rowe, Bullock, Polkey, & Morris, 2001). Thus it
seems plausible that the AMF may reflect computations that span across
multiple domains, a speculation we already put forth in the first AMF paper
(Pylkkanen & McElree, 2007) and have later refined (Pylkkanen, 2008;
Pylkkanen, Oliveri, et al., 2009). A more specific way to put this hypothesis
is that the AMF reflects semantic composition and that semantic composition
employs mechanisms also used in other domains. To test this, our on-goingresearch is aimed at assessing whether the AMF composition effects extend to
combinatory phenomena in nonlinguistic domains.
LOOKING AHEAD
Moving forward, here is what we would like to not see happen in the cognitive
neuroscience of semantics: for neuroscientists to ‘‘reinvent the wheel’’ when it
comes to cognitive models of semantic representation. We understand that
linguistic theory offers a tremendous amount of detail, which may seem
daunting to try to relate to neurobiological data. Thus it may be tempting to
flatten some of those distinctions when aiming to characterise the brain bases
NEUROSCIENCE AND SEMANTIC THEORY 17
Downloaded By: [University of Pennsylvania] At: 03:49 11 April 2011
of linguistic functions, as in, for example, Peter Hagoort’s model, where all
combinatory operations of language are treated as examples of a general
‘‘unification’’ function (Hagoort, 2005). We actually think it is quite possible
that the brain performs composition in various domains in rather similarways*this intuition is largely what drives our work on the domain generality
of the AMF*but we do not believe that a flat cognitive model is the right
kind of theory to guide us when designing experiments. If we do not look for
certain (theoretically well-motivated) distinctions in the brain, we certainly
will not find them, should they, in fact, be there.
To conclude, natural language meaning exhibits many fascinating
phenomena, most of which one is not consciously aware of without engaging
in the formal study of semantics. If theoretical semantics and the cognitiveneuroscience of semantics are not in communication with each other, we
will never understand how the brain achieves the astounding task of
comprehending and producing the meanings of human language in all their
glory.
Manuscript received 16 April 2010
Revised manuscript received 21 September 2010
First published online month/year
REFERENCES
Amodio, D., & Frith, C. (2006). Meeting of minds: The medial frontal cortex and social cognition.
Nature Reviews Neuroscience, 7, 268�277.
Anderson, S. W., Barrash, J., Bechara, A., & Tranel, D. (2006). Impairments of emotion and real-
world complex behavior following childhood- or adult-onset damage to ventromedial prefrontal
cortex. Journal of the International Neuropsychological Society, 12, 224�235.
Barker, C. (2002). Continuations and the nature of quantification. Natural Language Semantics, 10,
211�242.
Baron-Cohen, S., & Ring, H. (1994). A model of the mindreading system: Neuropsychological and
neurobiological perspectives. In P. Mitchell & C. Lewis (Eds.), Origins of an understanding mind
(pp. 183�207). Hove, UK: Lawrence Erlbaum.
Baron-Cohen, S., Ring, H., Moriarty, J., Schmitz, B., Costa, D., & Ell, P. (1994). The brain basis of
theory of mind: The role of the orbito-frontal region. British Journal of Psychiatry, 165, 640�649.
Barrash, J., Tranel, D., & Anderson, S. W. (2000). Acquired personality disturbances associated
with bilateral damage to the ventromedial prefrontal region. Developmental Neuropsychology,
18, 355�381.
Bechara, A., Damasio, H., & Damasio, A. R. (2000). Emotion, decision making and the
orbitofrontal cortex. Cerebral Cortex, 10, 295�307.
Bemis, D., & Pylkkanen, L. (2010). Simple composition in language processing: An MEG
investigation into the comprehension of minimal linguistic phrases. Manuscript submitted for
publication.
Berlin, H. A., Rolls, E. T., & Kischka, U. (2004). Impulsivity, time perception, emotion, and
reinforcement sensitivity in patients with orbitofrontal cortex lesions. Brain, 127, 1108�1126.
18 PYLKKANEN, BRENNAN, BEMIS
Downloaded By: [University of Pennsylvania] At: 03:49 11 April 2011
Brennan, J., Nir, Y., Hasson, U., Malach, R., Heeger, D., & Pylkkanen, L. (2010). Syntactic
complexity predicts anterior temporal activity during natural story listening. Brain & Language.
Advance online publication. doi: 10.1016/j.bandl.2010.04.002
Brennan, J., & Pylkkanen, L. (2008). Processing events: Behavioral and neuromagnetic correlates of
aspectual coercion. Brain & Language, 106, 132�143.
Brennan, J., & Pylkkanen, L. (2010). Processing psych verbs: Behavioral and MEG measures of two
different types of semantic complexity. Language and Cognitive Processes, 25(6), 777�807. doi:
10.1080/01690961003616840
Brennan, J., & Pylkkanen, L. (2010). The time-course and spatial distribution of brain activity
associated with processing sentences. Manuscript submitted for publication.
Burgess, P. W., & Wood, R. L. (1990). Neuropsychology of behaviour disorders following brain
injury. In R. L. Wood (Ed.), Neurobehavioural sequelae of traumatic brain injury (pp. 110�133).
New York: Taylor and Francis.
Caplan, D., Alpert, N., & Waters, G. (1999). PET studies of syntactic processing with auditory
sentence presentation. Neuroimage, 9, 343�351.
Damasio, A. R. (1994). Descartes’ error: Emotion, reason, and the human brain. New York: Putnam.
Davidson, D. (1967). The logical form of action sentences. In N. Reschler (Ed.), The logic of
decision and action (pp. 81�95). Pittsburgh: University of Pittsburgh Press.
Dehaene-Lambertz, G. (1997). Electrophysiological correlates of categorical phoneme perception
in adults. Neuroreport, 8(4), 919�924.
Dowty, D. R. (1979). Word meaning and Montague grammar. Dordrecht: Reidel.
Drenhaus, H., beim Graben, P., Saddy, D., & Frisch, S. (2006). Diagnosis and repair of negative
polarity constructions in the light of symbolic resonance analysis. Brain & Language, 96(3), 255�268.
Embick, D., Marantz, A., Miyashita, Y., O’Neil, W., & Sakai, K. L. (2000). A syntactic
specialization for Broca’s area. Proceedings of the National Academy of Sciences, 23, 6150�6154.
Frege, G. (1892). Uber sinn und bedeutung. Zeitschrift fur Philosophie und Philosophische Kritiki,
100, 25�50.
Friederici, A. D., Meyer, M., & von Cramon, D. Y. (2000). Auditory language comprehension: An
event-related fMRI study on the processing of syntactic and lexical information. Brain &
Language, 75(3), 465�477.
Gallagher, H. L., & Frith, C. D. (2003). Functional imaging of ‘theory of mind’. Trends in Cognitive
Science, 7(2), 77�83.
Grafman, J., Schwab, K., Warden, D., Pridgen, A., Brown, H. R., & Salazar, A. M. (1996). Frontal
lobe injuries, violence, and aggression: A report of the Vietnam head injury study. Neurology,
46, 1231�1238.
Grodzinsky, Y., & Friederici, A. D. (2006). Neuroimaging of syntax and syntactic processing.
Current Opinion in Neurobiology, 16, 240�246.
Hagoort, P. (2005). On Broca, brain, and binding: A new framework. Trends in Cognitive Sciences,
9, 416�423.
Hansen, P., Kringelbach, M., & Salmelin, R. (2010). MEG: An introduction to methods. Oxford:
Oxford University Press.
Heim, I., & Kratzer, A. (1998). Semantics in generative grammar. Oxford: Blackwell.
Hendriks, H. (1988). Type change in semantics: The scope of quantification and coordination. In
E. Klein & J. Van Benthem (Eds.), Categories, polymorphism and unification (pp. 96�119).
Amsterdam: ITLI.
Hickok, G., & Poeppel, D. (2000). Towards a functional neuroanatomy of speech perception.
Trends in Cognitive Sciences, 4, 131�138.
Humphries, C., Binder, J. R., Medler, D. A., & Liebenthal, E. (2006). Syntactic and semantic
modulation of neural activity during auditory sentence comprehension. Journal of Cognitive
Neuroscience, 18(4), 665�679.
NEUROSCIENCE AND SEMANTIC THEORY 19
Downloaded By: [University of Pennsylvania] At: 03:49 11 April 2011
Jackendoff, R. (2002). Foundations of language. Oxford: Oxford University Press.
Jackendoff, R. (1997). The architecture of the language faculty. Cambridge, MA: MIT Press.
Jacobson, P. (1999). Towards a variable-free semantics. Linguistics and Philosophy, 22, 117�184.
Kaan, E., & Swaab, T. Y. (2002). The brain circuitry of syntactic comprehension. Trends in
Cognitive Sciences, 6, 350�356.
Koenigs, M., & Tranel, D. (2007). Irrational economic decision-making after ventromedial
prefrontal damage: Evidence from the Ultimatum Game. Journal of Neuroscience, 27(4), 951�956.
Kratzer, A. (1996). Severing the external argument from its verb. In J. Rooryck & L. Zaring (Eds.),
Phrase structure and the lexicon (pp. 109�137). Dordrecht: Kluwer Academic.
Krueger, F., Barbey, A. K., & Grafman, J. (2009). The medial prefrontal cortex mediates social
event knowledge. Trends in Cognitive Sciences, 13, 103�109.
Kutas, M., & Hillyard, S. A. (1980). Reading senseless sentences: Brain potentials reflect semantic
incongruity. Science, 207, 203�205.
Ladusaw, W. (1979). Polarity sensitivity as inherent scope relations (Ph.D. dissertation, University of
Texas, Austin).
Ladusaw, W. (1980). On the notion ‘‘affective’’ in the analysis of negative polarity items. Journal of
Linguistic Research, 1, 1�23.
Ladusaw, W. (1983). Logical form and conditions on grammaticality. Linguistics and Philosophy, 6,
373�392.
Lau, E. F., Phillips, C., & Poeppel, D. (2008). A cortical network for semantics: (De)constructing
the N400. Nature Reviews Neuroscience, 9, 920�933.
Maguire, E. A., Frith, C. D., & Morris, R. G. (1999). The functional neuroanatomy of
comprehension and memory: The importance of prior knowledge. Brain, 122, 1839�1850.
Martin, A., & Chao, L. L. (2001). Semantic memory and the brain: Structure and processes.
Current Opinion in Neurobiology, 11, 194�201.
Mazoyer, B. M., Tzourio, N., Frak, V., Syrota, A., Murayama, N., Levrier, O., . . . Mehler, J. (1993).
The cortical representation of speech. Journal of Cognitive Neuroscience, 5(4), 467�479.
Montague, R. (1970). Universal grammar. Theoria, 36, 373�398.
Moro, A., Tettamanti, M., Perani, D., Donati, C., Cappa, S. F., & Fazio, F. (2001). Syntax and the
brain: Disentangling grammar by selective anomalies. NeuroImage, 13, 110�118.
Mummery, C. J., Patterson, K., Hodges, J. R., & Price, C. J. (1998). Functional neuroanatomy of
the semantic system: Divisible by what? Journal of Cognitive Neuroscience, 10, 766�777.
Naatanen, R., Lehtokoski, A., Lennes, M., Cheour, M., Huotilainen, M., Iivonen, A., . . . Conturo,
T. E. (1997). Anatomic localization and quantitative analysis of gradient refocused echo-planar
fMRI susceptibility artifacts. NeuroImage, 6, 156�167.
Nathaniel-James, D. A., & Frith, C. D. (2002). The role of the dorsolateral prefrontal cortex:
Evidence from the effects of contextual constraint in a sentence completion task. NeuroImage,
16, 1094�1102.
Nieuwland, M. S., Petersson, K. M., & Van Berkum, J. J. A. (2007). On sense and reference:
Examining the functional neuroanatomy of referential processing. NeuroImage, 37(3), 993�1004.
Ojemann, J. G., Akbudak, E., Snyder, A. Z., McKinstry, R. C., Raichle, M. E., & Conturo, T. E.
(1997). Anatomic localization and quantitative analysis of gradient refocused echo-planar
fMRI susceptibility artifacts. Neuroimage, 6(3), 156�167.
Parsons, T. (1990). Events in the semantics of English: A study of subatomic semantics. Cambridge,
MA: MIT Press.
Partee, B. H., & Rooth, M. (1983). Generalized conjunction and type ambiguity. In R. Bauerle,
C. Schwarze, & A. von Stechow (Eds.), Meaning, use, and interpretation of language (pp. 361�383). Berlin: Walter de Gruyter.
Patel, A. D. (2003). Language, music, syntax and the brain. Nature Neuroscience, 6, 674�681.
20 PYLKKANEN, BRENNAN, BEMIS
Downloaded By: [University of Pennsylvania] At: 03:49 11 April 2011
Pietroski, P. (2002). Function and concatenation. In G. Preyer & G. Peters (Eds.), Logical form and
language (pp. 91�117). Oxford: Oxford University Press.
Pietroski, P. (2004). Events and semantic architecture. Oxford: Oxford University Press.
Pietroski, P. (2006). Interpreting concatenation and concatenates. Philosophical Issues, 16, 221�245.
Pustejovsky, J. (1995). The generative lexicon. Cambridge, MA: MIT Press.
Pylkkanen, L. (2008). Mismatching meanings in brain and behavior. Language and Linguistics
Compass, 2(4), 712�738.
Pylkkanen, L., Martin, A. E., McElree, B., & Smart, A. (2009). The anterior midline field: Coercion
or decision making? Brain and Language, 108, 184�190.
Pylkkanen, L., & McElree, B. (2006). The syntax-semantic interface: On-line composition of
sentence meaning. In M. Traxler & M. A. Gernsbacher (Eds.), Handbook of psycholinguistics
(2nd ed., pp. 537�577). New York, NY: Elsevier.
Pylkkanen, L., & McElree, B. (2007). An MEG study of silent meaning. Journal of Cognitive
Neuroscience, 19, 1905�1921.
Pylkkanen, L., Oliveri, B., & Smart, A. (2009). Semantics vs. world knowledge in prefrontal cortex.
Language and Cognitive Processes, 24, 1313�1334.
Rogalsky, C., & Hickok, G. (2008). Selective attention to semantic and syntactic features modulates
sentence processing networks in anterior temporal cortex. Cerebral Cortex, 19, 786�796.
Rowe, A. D., Bullock, P. R., Polkey, C. E., & Morris, R. G. (2001). ‘Theory of mind’ impairments
and their relationship to executive functioning following frontal lobe excisions. Brain, 124, 600�616.
Saddy, D., Drenhaus, H., & Frisch, S. (2004). Processing polarity items: Contrastive licensing costs.
Brain & Language, 90, 495�502.
Schoenbaum, G., Roesch, M. R., & Stalnaker, T. A. (2006). Orbitofrontal cortex, decision-making
and drug addiction. Trends in Neuroscience, 29, 116�124.
Shamay-Tsoory, S. G., Tibi-Elhanany, Y., & Aharon-Peretz, J. (2006). The ventromedial prefrontal
cortex is involved in understanding affective but not cognitive theory of mind stories. Social
Neuroscience, 1, 149�166.
Shamay-Tsoory, S. G., Tomer, R., & Aharon-Peretz, J. (2005). The neuroanatomical basis of
understanding sarcasm and its relationship to social cognition. Neuropsychology, 19, 288�300.
Steedman, M. (1996). Surface structure and interpretation. Cambridge, MA: MIT Press.
Steinhauer, K., Drury, J. E., Portner, P., Walenski, M., & Ullman, M. T. (2010). Syntax, concepts,
and logic in the temporal dynamics of language comprehension: Evidence from event-related
potentials. Neuropsychologia, 48(6), 1525�1542.
Stowe, L. A., Broere, C. A., Paans, A. M., Wijers, A. A., Mulder, G., Vaalburg, W., & Zwaartz, F.
(1998). Localizing components of a complex task: Sentence processing and working memory.
Neuroreport, 9(13), 2995�2999.
Stromswold, K., Caplan, D., Alpert, N., & Rauch, S. (1996). Localization of syntactic
comprehension by positron emission tomography. Brain & Language, 52, 452�473.
Vandenberghe, R., Nobre, A. C., & Price, C. J. (2002). The response of left temporal cortex to
sentences. Journal of Cognitive Neuroscience, 14(4), 550�560.
Wallis, J. D. (2007). Neuronal mechanisms in prefrontal cortex underlying adaptive choice behavior.
Annals of the New York Academy of Sciences, 1121, 447�460.
Xiang, L., Dillon, B., & Phillips, C. (2009). Illusory licensing effects across dependency types: ERP
evidence. Brain & Language, 108, 40�55.
NEUROSCIENCE AND SEMANTIC THEORY 21
Downloaded By: [University of Pennsylvania] At: 03:49 11 April 2011
Top Related