ArticlePDF Available

Sentences in the Brain: Event-Related Potentials as Real-Time Reflections of Sentence Comprehension and Language Learning


Abstract and Figures

From the perspective of a person trying to understand a sentence, language is a continuous flow of information distributed over time. Somehow, the listener trans-lates this stream of information into discrete and rapidly sequenced units of sound, meaning, and structure, and does so in real time, that is, nearly instantaneously. The results of these complex analyses are then integrated into a single coherent inter-pretation, even in the presence of considerable ambiguity. The challenge facing us is to account for how people accomplish this. Asking the listener or reader for an answer to this question provides little useful information; the relevant processes are not generally available to conscious reflection. We must there-fore rely on other methods of investigation. One might surmise that the ideal method would mirror the properties of comprehension itself (Osterhout, McLaughlin, & Ber-sick, 1997). The ideal method should provide continuous measurement throughout the process of understanding a sentence, have a temporal resolution exceeding that of the relevant processes, and be differentially sensitive to events occurring at different levels of analysis (phonological, syntactic, semantic, etc.). Furthermore, language comprehension is clearly a function of the human brain. Theoretical accounts of some of the most fundamental questions about human language (e.g., the ability of children to acquire language, and the effects of age on language-learning ability) rely explicitly on biological explanations. Compelling tests of these hypotheses necessarily require biological measurement. For these and other reasons, the ideal method should also provide a means for relating the obtained data to brain function.
Content may be subject to copyright.
© 2004 by CRC Press LLC
Chapter 14
Sentences in the Brain:
Event-Related Potentials as
Real-Time Reflections of
Sentence Comprehension
and Language Learning
From the perspective of a person trying to understand a sentence, language is a
continuous flow of information distributed over time. Somehow, the listener trans-
lates this stream of information into discrete and rapidly sequenced units of sound,
meaning, and structure, and does so in real time, that is, nearly instantaneously. The
results of these complex analyses are then integrated into a single coherent inter-
pretation, even in the presence of considerable ambiguity.
The challenge facing us is to account for how people accomplish this. Asking the
listener or reader for an answer to this question provides little useful information; the
relevant processes are not generally available to conscious reflection. We must there-
fore rely on other methods of investigation. One might surmise that the ideal method
would mirror the properties of comprehension itself (Osterhout, McLaughlin, & Ber-
sick, 1997). The ideal method should provide continuous measurement throughout
the process of understanding a sentence, have a temporal resolution exceeding that of
the relevant processes, and be differentially sensitive to events occurring at different
levels of analysis (phonological, syntactic, semantic, etc.). Furthermore, language
comprehension is clearly a function of the human brain. Theoretical accounts of some
of the most fundamental questions about human language (e.g., the ability of children
to acquire language, and the effects of age on language-learning ability) rely explicitly
on biological explanations. Compelling tests of these hypotheses necessarily require
biological measurement. For these and other reasons, the ideal method should also
provide a means for relating the obtained data to brain function. Page 271 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
These properties of the ideal tool amount to a formidable methodological chal-
lenge. Unfortunately, few of the available methods have all of these properties, and
the absence of even one property can limit significantly the utility of the method.
For example, eyetracking provides continuous measurement during comprehension
and has been quite useful for modeling certain aspects of reading and spoken
language comprehension. However, eye movements respond similarly to events
occurring at different levels of linguistic analysis; for example, eye fixations and
regressions occur following an anomaly at any linguistic level. Furthermore, eye-
tracking does not provide any direct information about how language processing is
instantiated in the brain. Functional neuroimaging techniques (e.g., PET and fMRI)
do provide evidence concerning the relationship between language and brain and
have proven helpful in revealing the neurobiology of certain cognitive functions
(e.g., visual perception; cf. Tootell et al., 1998). However, nearly all of the currently
available imaging methods suffer from a temporal resolution (> 1 s) that is at least
an order of magnitude worse than the presumed temporal resolution of the processes
of interest (tens and hundreds of milliseconds). The consequences of this and related
limitations will be discussed in later sections of this chapter. (Further discussion of
the properties of these methods can be found in Mitchell, chapter 2, this volume.)
One method that approximates the ideal method is the recording of event-related
brain potentials (ERPs) elicited during language processing (Osterhout et al., 1997).
ERPs are scalp-recorded changes in electrical activity that occur in response to a
sensory, cognitive, or motor event (Rugg & Coles, 1995; van Berkum, chapter 13,
this volume). ERPs are thought to reflect the summed, simultaneously occurring
postsynaptic activity within neocortical pyramidal neurons. Topographical features
of the ERP are referred to as components and can be described in terms of polarity
(positive and negative), amplitude, onset, peak latency, and scalp distribution. ERPs
provide a millisecond-by-millisecond record of the brain’s electrical activity during
the process of interest. ERPs are multidimensional (varying in polarity, latency,
source configurations, etc.). Because they are direct reflections of brain activity,
ERPs offer the prospect of tying cognitive theories of sentence processing more
closely to the underlying neurobiology.
These prospective advantages are insignificant unless ERPs have the necessary
sensitivity to the processes of interest. The goal of this chapter is to selectively
review evidence (primarily from our laboratory) that ERPs are highly sensitive to
the relevant processes, and that this sensitivity can be exploited to illuminate the
cognitive and neurobiological underpinnings of sentence comprehension in native
speakers and in language learners. In particular, we will demonstrate how the unique
properties of ERPs permit novel tests of two fundamental tenets of modern-day
psycholinguistics: the belief that syntactic processing always precedes semantic
processing, and the belief that success at learning a language in adulthood is limited
by an age-related decline in brain plasticity. The ERP evidence we describe here
provides novel and perhaps unique perspectives on these widely held assumptions.
Ultimately, however, all of the available methods (including ERPs) are imperfect
tools. The most compelling conclusions often result from a convergence of evidence
coming from very different methods. Page 272 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
Contemporary linguistic theories posit a number of distinct and largely independent
levels of linguistic knowledge (Chomsky, 1986). A standard typology includes the
levels of phonology, syntax, and semantics. The task of a linguist is to describe the
units at each level and the rules that govern their combination. Psycholinguistic
theories need to specify how closely this typology, which is based on linguistic
observation and description, describes the cognitive and neurobiological events
underlying the comprehension of a sentence. Critically, however, the psycholinguist’s
task does not end with the description of a simple typology but must extend to an
account of how these different sources of knowledge are recruited in the service of
comprehending a sentence. This includes, for example, specifying the order in which
various sources of information are employed and the interactions (or lack of inter-
actions) between them. The ultimate and most challenging step is to describe how
these processes are implemented in the brain.
14.2.1Syntax and Semantics
Perhaps the most fundamental distinction is the one between sentence structure
(syntax) and sentence meaning (semantics). To many linguists, sentences that violate
syntactic constraints (such as The cat will ate the food) are clearly distinct from
sentences that violate semantic constraints (such as John buttered his bread with
socks). Whether this distinction characterizes accurately the processes underlying
sentence comprehension has been a matter of debate. A standard assumption under-
lying much psycholinguistic work is that a relatively direct mapping exists between
the levels of knowledge posited within linguistic theories and the cognitive and
neural processes underlying comprehension (Bock & Kroch, 1989; Fodor, 1983;
Forster, 1979). Distinct language-specific processes are thought to interpret a sen-
tence at each level of analysis, and distinct representations are thought to result from
these computations. However, other theorists, most notably those working in the
neural net modeling domain, deny that this mapping exists (Elman, 1990;
McClelland, St. John, & Taraban, 1989). Instead, the meaning of the sentence is
claimed to be derived directly, without an intervening level of syntactic structure.
This debate is mirrored in the language dysfunction literature. The syndromes
known as Broca’s and Wernicke’s aphasia were initially given motor and sensory
interpretations, respectively. This approach gave way to linguistic descriptions in which
each disorder was claimed to reflect the loss of a specific type of linguistic information
or function (Caramazza & Zurif, 1976). Broca’s aphasia was claimed to manifest (in
addition to other impairments) an impairment in syntactic function. Wernicke’s aphasia
was associated with a degradation in the access to lexical semantics. These “linguistic”
accounts of aphasia not only accepted the distinction between sentence form and
meaning but also localized these aspects of language processing in different parts of
the brain. Syntactic processes were claimed to be located in restricted parts of the left
inferior frontal lobe, whereas semantic processes were claimed to be located in the
posterior portion of the left temporal lobe. However, although this account has been Page 273 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
quite influential, a closer inspection of the aphasic syndromes raises questions. The
correlations between lesion site and the linguistically described aphasic symptoms turn
out to be weak (Caplan, Hildebrandt, & Makris, 1996; Dronkers, 2000), and the aphasic
symptoms tend to vary dramatically within a person over time (McNeil & Doyle,
2000). Such considerations have led some researchers to question the accuracy and
utility of linguistically based accounts of aphasia (Bates & Dick, 2000; Blumstein &
Milberg, 2000; McNeil & Doyle; Milberg & Blumstein, 2000).
14.2.2 Neuroimaging and the Syntax/Semantics
Neuroimaging methods provide another means for identifying the types of processes
used in sentence comprehension and mapping them onto the brain (Fiebach et al.,
chapter 17, this volume). For example, fMRI allows researchers to measure changes
in blood flow (usually by recording changes in the blood oxygenation level dependent
[BOLD] response; Jezzard, Matthews, & Smith, 2001) while neurologically intact
subjects read or listen to tokens of the language. It is assumed that the BOLD
response indirectly reflects neural activity. Although these methods represent major
advances for the field of cognitive neuroscience, they are not without complications
as tools for studying real-time language comprehension. First, the hemodynamic
response to an event is delayed several seconds and evolves over 10 to 15 s. This
temporal resolution contrasts starkly with the fact that in normal fluent conversation,
speakers produce (on average) three words, four syllables, and 12 phonemes per sec
(Levelt, 1999). Furthermore, the processing of a single linguistic unit, for example
a word, most likely involves a constellation of processes, each having temporal
durations of considerably less than 1 s. In other words, under conditions that approx-
imate normal speaking and reading, it is very difficult to isolate the hemodynamic
response to a particular word, much less the (phonological, syntactic, semantic, etc.)
processing steps that occur in the processing of that word. In imaging experiments,
all of these processes are represented in the time dimension by a single data point.
Disentangling them (i.e., associating a specific process with a specific activated area)
is not a trivial matter.
Second, because language comprehension is inherently an integrative process,
one cannot reasonably assume that successive words and sentences are processed
independently. This fact complicates efforts to isolate the response to particular
words in a sentence by using event-related fMRI designs (e.g., Burock, Buckner,
Woldorff, Rosen, & Dale, 1998). Event-related designs measure the BOLD response
to rapidly sequenced individual events and assume that temporally overlapping
BOLD responses summate linearly. Although the independence of overlapping
hemodynamic functions has been demonstrated for simple visual stimuli (Dale and
Buckner, 1997), the same cannot be assumed for words in sentences.
Third, even if the BOLD response to a particular word in a sentence could be
isolated successfully, the BOLD response is maximally sensitive to sustained brain
events and relatively insensitive to transient events (Savoy et al., 1995). Conse-
quently, sustained brain events will always dominate the BOLD response and poten-
tially obscure the transient language-related events of interest.1
PE: Year
lomitted as
per APA
style that it is
not needed
for subse-
quent cita-
tion within a
para. Page 274 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
These complications notwithstanding, fMRI has been used successfully to eluci-
date some aspects of word and sentence processing. The most fundamental and
consistent finding has been that words and sentences activate the perisylvian areas
of the left hemisphere that are classically associated with language dysfunction
following brain damage (Bavelier, Corina, & Jezzard et al, 1997; Price, 1998). How-
ever, it is not unusual for areas in addition to these classical areas to show equally
strong activation, and not all of the relevant perisylvian areas are always activated.
The cross-experiment variability in fMRI results increases substantially when
specific linguistic processes are examined. For example, studies attempting to local-
ize semantic processing have reported activated areas including most of the left (and
sometimes right) temporal lobe and various parts of the frontal lobe (Binder & Price,
2001). Attempts to localize syntactic processes have also varied widely, and have
included the left inferior frontal cortex (Broca’s area and surrounding cortex) and
parietal and temporal regions, sometimes but not always lateralized to the left
Undoubtedly, some of this variation occurs due to differences in stimuli, task,
and procedures. It is therefore instructive to examine results obtained in a relatively
homogeneous set of studies that employ a common strategy for isolating a particular
linguistic process. One seemingly straightforward stimulus manipulation contrasts
the BOLD response to anomalies that involve different linguistic levels, for example,
syntactic and semantic; the assumption is that the anomaly will engender more
processing at the linguistic level of the anomaly, and that this processing will be
reflected in a larger BOLD response in the affected areas of the brain. At least six
neuroimaging studies employing this strategy have been published recently
(Friederici, Ruschemeyer, Hahne, & Fiebach, 2003; Kang, Constable, Gore, &
Avrutin, 1999; Kuperberg et al., 2000; Kuperberg, Holcomb, et al., 2003; Newman,
Pancheva, Ozawa, Ozawa, Neville, & Ullman, 2001; Ni et al., 2000). Perhaps the
most striking aspect of these collective results is the variability in outcome and
conclusion (see Kuperberg, Holcomb, et al., for an excellent discussion of this point).
For example, Kuperberg et al. (2000) conclude that the left and right temporal lobes
are involved in semantic processing but do not identify any areas that are modulated
specifically by syntactic anomalies. Ni et al. and Kang et al. conclude that the left
inferior frontal lobe is involved in syntactic processing, whereas the posterior tem-
poral lobes are more involved in semantic processing. Newman et al. (2001) conclude
that superior frontal regions are involved in syntactic processing and inferior frontal
and temporal regions are involved in semantic processing. In the two most recent
studies, Kuperberg et al. (2003) conclude that overlapping networks (including
inferior frontal and left temporal regions) are modulated in opposite directions by
the two types of anomaly, with semantic anomalies evoking a larger BOLD response
in this area and syntactic anomalies a weaker response, relative to a nonanomalous
sentence condition. Friederici et al. (2003) conclude that syntactic anomalies evoke
more activity in inferior frontal and anterior temporal regions in the left hemisphere,
whereas semantic anomalies evoke activity bilaterally in the midtemporal regions.
There are a few notable consistencies across these studies. For example, the
authors are in agreement (except Kuperberg, Holcomb, et al., 2003) that their syn-
tactic and semantic manipulations identified distinct regions of the brain as being
AU: Provide
all author
names. Page 275 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
more involved in one level of analysis than the other. However, there is considerable
disagreement about which brain regions are involved in syntactic and semantic
processing. It is reasonable to speculate that this variance in outcome is due, in part,
to the difficulty of isolating specific types of linguistic processes using fMRI when
these processes are transient and closely sequenced in time.2
14.2.3ERPs and the Syntax/Semantics Distinction
Unlike fMRI, ERPs have a temporal resolution exceeding that of the processes
underlying language comprehension. ERPs are also maximally sensitive to transient
events occurring in the brain and much less affected by sustained events. These
advantageous properties allow one to hope that ERPs might be more successful at
isolating individual processes as they occur in real time during sentence compre-
This hope has been fulfilled by two decades of accumulating evidence, much of
it investigating the brain responses to syntactic and semantic anomalies. The crucial
finding is that syntactic and semantic anomalies elicit qualitatively distinct ERP
effects, and that these effects are characterized by distinct and consistent temporal
properties. Semantic anomalies (e.g., The cat will bake the food … ) elicit a negative
wave that peaks at about 400 ms after the anomalous word appears (the N400 effect;
Figure 14.1A) (Kutas & Hillyard, 1980, 1984; Osterhout & Nicol, 1999). By contrast,
syntactic anomalies (e.g., The cat will eating the food … ) elicit a large positive
wave that onsets at about 500 ms after presentation of the anomalous word and
persists for at least half a second (the P600 effect; Figure 14.1B) (Hagoort, Brown,
& Groothusen, 1993; McKinnon & Osterhout, 1996; Osterhout, 1997; Osterhout &
Holcomb, 1992, 1993; Osterhout, Holcomb, & Swinney, 1994; Osterhout & Mobley,
1995; Osterhout, McKinmon, Bersick, & Corey, 1996; Osterhout, McLaughlin,
Allen, & Inoue, 2002; Osterhout & Nicol, 1999). In some studies, syntactic anom-
alies have also elicited a negativity over anterior regions of the scalp, with onsets
ranging from 100 to 300 ms (Friederici, 1995; Neville, Nicol, Barss, Forster, &
Garret, 1991; Osterhout & Holcomb, 1992; Osterhout & Mobley, 1995). These
results are highly reproducible3 and generalize well across types of anomaly (with
anomalies involving phrase structure, agreement, verb subcategorization, and con-
stituent-movement all eliciting P600-like effects), types of languages (including
word-order languages such as English, Dutch, and French, and case-marked lan-
guages such as Italian and Japanese; Angrilli et al., 2002; Hagoort et al., 1993; Inoue
& Osterhout, in preparation; Nakagome et al., 2001), and various methodological
factors (including modality of the input, rate of word presentation, and presenting
isolated sentences and natural prose; Allen, Badecker, & Osterhout, in press;
McKinnon & Osterhout, 1996; Osterhout & Holcomb, 1993; Osterhout et al., 2002).
These robust effects seem to indicate that the brain does in fact honor the distinction
between the form and the meaning of a sentence.
Furthermore, the available evidence suggests that the N400, but not the P600,
is sensitive to a wide range of word properties, including word frequency, morpho-
logical content, and semantic relatedness. Conversely, the P600, but not the N400,
is sensitive to the syntactic well-formedness of the sentence. A recent study by Allen,
AU: Not
included in
ref list. AU: Pub-
lished yet? Page 276 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
et al. (in press) nicely illustrates this point. Allen et al. independently manipulated
the normative frequency and syntactic well-formedness of verbs in sentences. These
Figure 14.1. (A) ERPs (recorded over the vertex) elicited by semantically anomalous words
(dashed line) and nonanomalous control words (solid line) in sentences such as The cat will
eat/bake the food. (B) ERPs elicited by syntactically anomalous (dashed line) and nonanom-
alous control words (solid line) in sentences such as The cat will eat/eating the food. Onset
of the critical word is indicated by the vertical bar. Each hashmark represents 100 ms The
vertical calibration bar represents 5 µV. Adapted from “On the distinctiveness, independence,
and time course of the brain responses to syntactic and semantic anomalies,” by L. Osterhout
and J. Nicol, 1999, Language and Cognitive Process, 14, 283–317.
200 400 600 800
The cat will EAT
The cat will BAKE
200 400 600 800
The cat will EAT
*The cat will EATING
AU: Pub-
lished yet? Page 277 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
factors had independent effects: Word frequency affected N400 amplitude but not
P600 amplitude, and syntactic well-formedness affected P600 amplitude but not
N400 amplitude (Figure 14.2).
All of this ERP evidence converges on the conclusion that separable lexico-
semantic and syntactic processes exist, and that manifestations of this distinction
can be observed in scalp-recorded electrical brain activity. Furthermore, the ERP
evidence shows quite clearly that the syntactic and semantic (or relevant correlated)
processes are temporally distinct and transient events—and that ERPs, unlike fMRI,
are capable of temporally isolating these (or correlated) processes as they occur in
real time. That is, at a predictable moment in time during comprehension, the ERP
waveform is sensitive specifically to lexico-semantic manipulations, whereas at
another moment in time the waveform is sensitive specifically to syntactic manip-
Ideally, one would like to discern the neural sources of these language-sensitive
ERP effects. By doing so, one might be able to isolate the processes of interest in
both time and space. Unfortunately, the source of a given ERP effect (known as the
“inverse solution”) cannot be known with certainty. This follows from the fact that
a large number of source configurations could produce an identical pattern of activity
across the scalp (Nunez, 1995). Nonetheless, source estimates are possible, given
certain limiting assumptions. The traditional approach to source localization has
been to search for point dipole sources (Hämäläinen & Sarvas, 1989; Henderson,
Butler, & Glass, 1975; Kavanaugh, Darcey, Lehmann, & Fender, 1978). In general,
this entails assuming a small number of dipole sources and iterating through all
possible combinations of dipole location, orientation, and strength, looking for the
best match between the source model and the observed scalp distribution. This
method brings with it numerous limitations and caveats (Halgren et al., 2002).
More recently developed alternative methods provide a true tomographic anal-
ysis analogous to that provided by neuroimaging methods. For example, Low Res-
olution Electromagnetic Tomography (LORETA; Pascual-Marqui, Michel, & Leh-
mann, 1994) estimates the current distribution throughout the entire three-
dimensional brain. The primary assumption of this approach is that dramatic changes
in current do not occur across contiguous areas of cortex (i.e., in adjacent voxels).
The primary limitation is that the maximum spatial resolution is perhaps as low as
1 cm3, significantly worse than fMRI or PET. The primary advantage, a crucial one,
is that LORETA can provide an estimate of current distribution for each sample of
brain activity (i.e., one estimate every few ms). Although LORETA should be viewed
as a model or estimate rather than as an observation of current distribution, its
reliability and validity have been established in some nonlinguistic domains
(Anderer, Pascual-Marqui, Smlitsch, & Saletu, 1998; Pascual-Marqui, 1999; Pas-
cual-Marqui, Esslen, Kochi, & Lehmann, 2002).
In our lab, we have been using LORETA to provide estimates of current distri-
bution for the peaks of the N400 and P600 effects. In one study, we recorded ERPs
from 20 native English speakers while they read sentences that were well-formed
and coherent, or contained a syntactic or semantic anomaly. As expected, the syn-
tactic and semantic anomalies elicited P600 and N400 effects, respectively. We then
performed the LORETA analysis on the N400 peak (in both the semantically well- Page 278 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
formed semantically anomalous conditions) and on the P600 midpoint in the
syntactically anomalous condition, and on the corresponding sample in the syntac-
tically well-formed condition.4
Figure 14.2. (A) ERPs (recorded over the vertex) to high-frequency (solid line) and low-
frequency (dashed line) verbs in grammatical sentences such as The man will work/sway on
the platform. (B) ERPs to high-frequency (solid line) and low-frequency (dashed line) verbs
in ungrammatical sentences such as The man will worked/swayed on the platform. (C) ERPs
to grammatical (solid line) and ungrammatical (dashed line) high-frequency verbs. (D) ERPs
to grammatical (solid line) and ungrammatical (dashed line) low-frequency verbs. Note that
the effects of word frequency and grammaticality have independent effects on the ERP.
Adapted from “Morphological analysis during sentence processing,” by M. D. Allen, W.
Badecker, and L. Osterhout, in press, Language and Cognitive Process.
HF Grammatical “... will work ...”
LF Grammatical “... will sway ...”
HF Grammatical “... will work ...”
LF Ungrammatical “... will worked ...”
LF Grammatical “... will sway ...”
LF Ungrammatical “... will swayed ...”
HF Ungrammatical “... will worked ...”
LF Ungrammatical “... will swayed ...”
A. Cz
B. Cz
C. Cz
D. Cz
200 400 600
AU: “and”? Page 279 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
The results of the LORETA analysis were striking. In the semantic condition
(Figure 14.3), the estimated current distribution for the N400 peak (at 410 ms) was
largest in the posterior middle temporal lobe and angular gyrus of the left hemisphere
[Brodmann’s area (BA) 39]. This was true for both the well-formed and semantically
anomalous conditions, although the current distribution in this area had greater
intensity in the anomalous condition. In the syntactic condition (Figure 14.4), the
primary source of current for the P600 midpoint (at 603 ms) was in the left inferior
frontal cortex (BA 44, 45, and 47). When the stimulus was syntactically anomalous,
the current spread throughout the left prefrontal cortex and into the left anterior
temporal lobe.
Our LORETA results are striking because they are highly consistent with the
classic lesion-based model of language and brain. The putative source of the N400
component, which is sensitive to properties of words and to meaning relations
between words, is located in the part of the brain that, when damaged, produces the
most problems with words and meaning (e.g., Dronkers, 2000). The putative source
of the P600 effect (and the same time point in the syntactically well-formed condi-
tion) is localized to the left inferior frontal cortex, which is classically associated
with agrammatism, and to the left anterior temporal cortex, which is the area most
Figure 14.3. LORETA solutions for N400 peak (at 410 ms) elicited by critical words in the
well-formed (top panel) and semantically anomalous (bottom panel) condition. See text for
more detail.
Semantically Well-formed
Semantically Ill-formed Page 280 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
commonly damaged in cases of agrammatism (Dronkers). These results are also
consistent to a degree with fMRI results. The posterior temporal lobe/angular gyrus
region of the left hemisphere (and in particular BA 39) is one of the few regions
that is almost always activated in studies attempting to isolate semantic processing
(Price, 1998). The inferior frontal lobe of the left hemisphere is frequently activated
in studies attempting to isolate syntactic processing (Caplan et al., 1996).5
One important caveat is that although the stimulus manipulations affecting the
N400 and P600 are well understood, the specific cognitive processes underlying
them are not. These effects might be direct manifestations of the syntactic and
semantic processes or they might be manifestations of processes that are correlated
with, but indeterminately removed from, the linguistic process themselves. Deciding
which of these possibilities is correct might turn out to be an intractable problem,
as impossible as altering the temporal properties of the brain’s hemodynamic
response or discovering perfect correlations between a lesion site and deficit patterns.
However, what seems to be intractable might become less so given data from other
methods of investigation. For example, given both the LORETA estimates of the
current distribution for the N400 and P600 effects and the lesion data, one could
rationally argue that these effects do in fact index specifically semantic and syntactic
Figure 14.4. LORETA solutions for the P600 midpoint (at 603 ms) elicited by critical words
in the well-formed (top panel) and syntactically anomalous (bottom panel) conditions.
Syntactically Well-formed
Syntactically Ill-formed
AU: Citation
in Note 5:
Dale et al.
year is 2000
in ref. list. Page 281 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
processing. Similarly, given both the highly variable semantic-processing fMRI
results and the lesion and ERP results, compelling claims can be made concerning
the regions in the brain most implicated in semantic processing. Although an appre-
ciation for the value of converging evidence is an old and perhaps out-of-fashion
virtue, its value has only increased with the advent of these new, but imperfect,
investigative methods.
14.2.4 Coordination Between Syntax and Semantics in
Sentence Processing
Because of their unique constellation of properties, ERPs can also be used to address
longstanding issues concerning the coordination of syntactic and semantic process-
ing. Theory regarding this issue has been deeply influenced by a family of syntax-
first models of linguistic structure (Chomsky, 1981) and language processing (Fer-
reira & Clifton, 1986; Fodor & Ferreira, 1998). The psycholinguistic models posit
that language comprehension is controlled by an initial stage of purely syntactic
processing. As words arrive in the linguistic input, they are rapidly organized into
a structural analysis by a process that is not influenced by detailed lexical or semantic
knowledge. The output of this syntactic process then guides semantic interpretation.
Syntax-first processing has been implicated in accounts of garden-path phenomena,
in which readers and listeners initially misanalyze sentences containing temporary
syntactic ambiguities. Consider, for example, sentences (1a) and (1b):
(1) a. The doctor believed the patient was lying.
b. The doctor believed the patient after hearing the story.
Readers of sentences like (1a) often initially interpret the second noun phrase
(the patient) as the direct object of the first verb (believed), when it is actually the
subject of an embedded clause (the patient was lying). Misanalysis is detected in
the form of processing difficulty when readers encounter was, which is only con-
sistent with the correct analysis. This processing difficulty is manifested in eye-
fixation times, regressive eye movements, and P600 effects in the ERP (Ferreira &
Clifton, 1986; Osterhout & Holcomb, 1992). Syntax-first models have claimed that,
due to the control of initial interpretative commitments by exclusively syntactic
mechanisms, garden-path errors occur even in the presence of potentially disambig-
uating semantic information (e.g., Ferreira & Clifton). In situations of temporary
syntactic ambiguity such as (1a, b), processing is controlled by a bias to choose the
simplest grammatically licensed structural analysis [in (1a, b), the direct object
analysis]. Whenever the simplest structural analysis is not the correct analysis,
garden-path effects are predicted.
While garden-path effects have been widely observed, other studies have shown
that they can be mitigated or eliminated by some types of nonsyntactic information.
For instance, Garnsey, Pearlmutter, Myers, & Lotocky, (1997) found that semantic
plausibility affected the likelihood of garden-path effects. Sentences in which the
direct-object interpretation is implausible (e.g., The doctor believed the medication
would work) result in small or nonexistent garden-path effects. Other studies have Page 282 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
shown effects of a wide range of constraints, including detailed lexico-syntactic
knowledge, semantics, and discourse knowledge (cf. MacDonald, Pearlmutter, &
Seidenberg, 1994; Trueswell & Tanenhaus, 1994). The influence of such factors on
ambiguity resolution seems inconsistent with the syntax-first prediction that initial
processing commitments are not influenced by nonsyntactic knowledge.
To account for these interacting influences of syntactic and nonsyntactic infor-
mation on sentence processing, theorists have developed a diverse family of
constraint-based models. These models posit a probabilistic constraint-satisfaction
process in which syntactic knowledge is only one of a number of constraints on
interpretation. Syntax-first models have not, however, been completely invalidated
by demonstrations of rapid interactivity. The most recent proposals continue to posit
an initial stage of purely syntactic processing, albeit one with a short duration
(Frazier & Clifton, 1996). According to these proposals, a rapid influence of non-
syntactic knowledge on sentence parsing can be explained in terms of re-analysis.
For example, if processing difficulty at was in (1a) is eased by semantic knowledge,
this might mean that a direct-object analysis was initially pursued but rapidly revised
using information about plausibility.
Thus, a consensus has emerged concerning the rapid nature of interaction
between syntactic and semantic knowledge, without resolving fundamental disagree-
ments about the degree to which syntactic processing controls other aspects of
language processing. These theoretical developments have brought the field against
some limitations of the standard garden-path paradigm. Garden-path experiments
examine the impact of nonsyntactic information exclusively in situations where
syntactic cues are indeterminate. The question is usually whether or not syntactic
processing is influenced by some nonsyntactic form of knowledge (or whether it is
However, the implicit assumption in all of this work (regardless of theoretical
bias) is that unless syntactic cues are indeterminate, syntax always controls the
direction of processing. We describe here a recent study in our laboratory (Kim,
Relkin, Lee, & Hirsc, in preparation) that examined semantic influences on sentence
comprehension in syntactically unambiguous situations. The goal of the study was
to test the widespread assumption that, in the absence of syntactic uncertainty,
syntactic information is “in charge” of sentence processing or, more specifically,
that semantic processing is always fundamentally dependent on the output of the
syntactic processing system. Kim et al. recorded ERPs while participants read strings
containing linguistic violations, as in (2a), and well-formed controls like (2b) and
(2) a. The mysterious crime had been solving the detective. (verb violation)
b. The mysterious crime had been solved by the detective. (passive control)
c. The brilliant detective had been solving mysterious crimes for decades.
(active control)
In (2a), the first noun phrase mysterious crime is highly plausible as the logical
object (theme) of the verb solve but anomalous as the logical subject (agent). The
syntactic cues in the sentence conflict directly with these semantic properties. The
AU: 1997 in
ref list. Page 283 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
-ing inflection of the verb is consistent with the anomalous agent interpretation but
not the theme interpretation [as opposed to -ed as in (2b)]. Sentence (2a) provides
a special situation—contrasted with the well-formed (2b) and (2c)—in which syn-
tactic and semantic constraints point toward directly opposed functional outcomes.
The interaction of syntactic and semantic processing in such situations can be studied
with ERPs. If syntactic processing controls semantic processing, then an agent
interpretation of the first noun phrase should be pursued. The implausibility of this
interpretation should elicit an enhanced N400 component for violation verbs (2a)
relative to control verbs (2b) and (2c). By contrast, semantic processing may operate
with some independence from syntactic control such that the plausible theme inter-
pretation of the noun phrase is pursued, even though it directly conflicts with
syntactic cues in the string. Because this interpretation is in direct conflict with the
syntactic cues in the string, it is possible that readers will encounter syntactic
processing difficulty at the verb. That is, powerful semantic cues may cause a well-
formed string to appear ill-formed. Such a perceived syntactic incongruity might
elicit a P600 effect. Note that the two different functional outcomes mentioned above
cannot be distinguished by reading time or eyetracking measures, which conflate
syntactic and semantic processing difficulty at the anomalous verb.
ERPs to the verbs in each sentence type are shown in Figure 14.5. Violation
verbs elicited a robust P600 effect compared to the control conditions, but no increase
in N400 amplitude. These results seem to indicate that syntactic processing difficulty
occurred at the verb when its inflection conflicted with available semantic cues. At
the same time, there is no indication that semantic processing is guided by syntactic
cues into the difficult logical-subject interpretation.6,7
The idea that syntactic processing precedes and drives semantic interpretation
is pervasive in psycholinguistics. Our results clearly do not fit well with that idea.
Instead, it appears that at least under some circumstances, semantic processing
exhibits a degree of independence from syntactic control and in fact seems to operate
in advance of, and drive, syntactic processing. An important theoretical issue raised
by this and related work (see Kolk, Chwilla, van Herten, & Oor, 2003; Kuperberg,
Sitnikova, Caplan, & Holcomb, 2003) is the notion of processing independence.
With respect to the claims of syntax-first models, the Kim et al. results indicate that
the independence of semantic processing is greater than predicted and that the
independence of syntactic processing is less than predicted.
We suggest that syntactic and semantic processing are indeed separated by a
form of independence, but a weak form. The independence of these processes enables
them to pursue an internally attractive analysis even when it is inconsistent with the
output of other processes. Furthermore, rather than operating in a serial fashion (as
proposed by the two-stage garden-path theory), syntactic and semantic processing
operate in parallel. Finally, it seems likely that these processes undergo a near-
constant interaction. Thus, syntactic processing might guide semantic processing in
John was disliked by Mary, where syntactic information points to the proper thematic
role assignments (i.e., the passive morphology indicates that John is the theme and
not the agent of dislike). Semantic processing might guide syntactic processing in
We gave John the money, where semantic knowledge supports the proper syntactic
AU: Note 7
not cited in
text. Placed
here only for
the purpose
of this note. Page 284 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
analysis (indirect vs. direct object) of John (i.e., John would make an excellent
recipient and an unlikely theme for the verb give).
These considerations suggest a potentially rich area of investigation concerning
the nature of syntactic and semantic processing. We know that in The mysterious
crime had been solving, the verb solving elicits a P600 effect and not an N400 effect.
Presumably, this happens because the syntactic analyzer tries to find an analysis
consistent with a “Predicate = solve, Theme = crime” interpretation. But we also
know that in The cats will bake the food, the verb bake elicits an N400 effect and
not a P600 effect (Osterhout & Nicol, 1999, and discussed in the “Syntax and
Semantics” section, this chapter). It appears that the implausibility of cats acting as
agents in this scenario does not cause the syntactic analyzer to flail away trying to
come up with an alternative structure; instead, an implausible but syntactically
supported interpretation is pursued, with difficulty.
The above observations raise a compelling paradox. The crime had been solving
and The cats will bake are quite similar; in each scenario, the subject is an implausible
Agent for the verb. Why, then, do the verbs in these two sentences elicit such different
brain responses? Two hypotheses come to mind. Perhaps the critical factor is that a
strong “semantic attraction” exists between the verb solve and the theme crime,
whereas no such attraction exists between the verb bake and cat. Alternatively, the
critical factor might be that the syntax of The mysterious crime had been solving
can be easily “fixed” to support a semantically plausible interpretation (simply
changing the verb’s inflection from -ing to -ed), whereas no such easy fix is available
Figure 14.5. ERPs (recorded over the posterior site PZ) elicited by critical words in the two
nonanomalous control conditions and in the anomalous verb violation condition (small
5 µv
200 400 600 800 1000
Verb Active Control
Verb Passive Control
Verb Violation Page 285 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
in the sentence The cats will bake the food. These two hypotheses can be contrasted
by examining sentences containing a strong semantic attraction, one that is incon-
sistent with the syntax and that cannot be made to fit the syntax with an easy syntactic
fix. For example, the sentence fragment The meal began to devour does not allow
an easy fix of the syntax to conform to a “theme = meal” interpretation, but the
system might try to find one anyway due to the semantic attraction between the verb
devour and the theme meal. If so, then devour should elicit a P600 effect, rather
than an N400 effect. Exactly this type of result has been reported in two recent
studies (Kolk et al., 2003; Kuperberg, Sitnikova, et al., 2003). Kuperberg, Sitnikova,
et al. report that the verb eat in the sentence For breakfast, the eggs began to eat
elicits a P600 effect, rather than an N400 effect. Apparently, strong semantic attrac-
tions between predicates and arguments induce the syntactic analyzer to work over-
time, searching for a structure that is consistent with the attraction. To phrase it
differently: The system commits itself to the semantic attraction (hence the lack of
an N400 effect) and insists on trying to “fix” the syntax (hence the presence of a
P600 effect).
Obviously, more work is needed to test the plausibility of this model and to
more completely understand how and when syntactic and semantic processing influ-
ence each other. What is already clear, however, is that ERPs are useful tools for
investigating complex issues in sentence comprehension that have been difficult to
study using other methods.
One of the standard tenets of modern-day psycholinguistics is that the ability to
learn a new language degrades with age. Age-related decreases in language-learning
ability have been reported for most aspects of language (including phonemes, words,
and grammar), and adults are often claimed to achieve lower levels of second-
language proficiency than children. Although the causes of these age effects on
proficiency are controversial, two frequently cited theoretical explanations stand out.
One explanation involves the putative existence of a critical period for second-
language learning. According to this explanation, language learning is constrained
by maturational factors (specifically, brain maturation) that circumscribe a critical
period (which, according to most accounts, ends around puberty) for native-like
attainment. Behavioral studies appear to confirm this notion by showing that the
proficiency scores of L2 learners on a variety of language tasks decline with the age
of initial exposure to the second language (Johnson, 1992; Johnson & Newport,
Another explanation attributes the age-related declines in L2 learning to the
effects of increasing experience with a first language. Neural network simulations
provide a convenient framework for understanding effects of L1 learning on L2
learning. In these simulations, early learning results in the entrenchment of optimal
network patterns, after which new learning requires considerable training. Consistent
with this view, it has been demonstrated that early experience with phonemes (Kuhl, Page 286 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
2000) and words (Ellis & Lambon Ralph, 2000) seems to degrade the ability to learn
new phonemes and words later in life.
Although on the surface these two explanations appear to be quite distinct, they
both implicate the same underlying cause for the age-of-acquisition effects on L2
learning, namely, a reduction in neural plasticity that degrades the ability to learn
new linguistic information. However, despite a considerable consensus about the
accuracy of this claim, there is little direct evidence to support it. The ideal method
for investigating the plasticity hypothesis would provide direct measurements of
brain responses to L2 words and sentences. This would permit researchers to evaluate
changes in brain activity that occur over time as a person attempts to acquire a new
Recently, several studies using neuroimaging methods (i.e., PET, MRI) have
investigated the reduction “in plasticity” hypothesis by contrasting groups of L2
speakers who acquired the L2 at different ages. A result indicating that, for example,
prepubescent learners represent their L1 and L2 similarly, whereas postpubescent
learners do not, would be consistent with the critical-period hypothesis. Exactly that
type of evidence was reported in one recent study (Kim et al., 1997). Early L2
learners showed highly similar patterns of activation for their L1 and L2, whereas
late learners showed differences, most notably in Broca’s area (for other studies
showing L1/L2 differences see Dehaene et al., 1997; Halsband, Krause, Sipila, Teras,
& Laihinen, 2002; Perani et al., 1996; Waterburger et al., 2003). However, other
studies have reported no significant effects of age of acquisition (Chee, Caplon et
al., 1999a; Chee, Tan et al., 1999b; Hasegawa, Capenter, & Just, 2002; Hernandez,
Martinez, & Kohnert, 2000; Illes et al., 1999; Klein, Milner, Zatorre, Zhao, &
Nikelski, 1999).
One partial explanation for the inconsistent results implicates the almost univer-
sal use of cross-sectional, between-subject designs in prior fMRI (and ERP) work
on L2 learning. This design makes it very difficult to properly control numerous
subject variables (e.g., L2 fluency and the quantity and quality of L2 experience)
that could easily confound the age effects (Chee et al., 2001; Perani et al., 1998).
The best way of avoiding such pitfalls is to longitudinally study a group of novice
learners as they become increasingly proficient in the L2. A longitudinal design
allows one to examine changes in brain activity and behavior relative to responses
observed during an initial state of near-total incompetence in the L2 within the same
group of learners. By including a group of subjects with native proficiency, one can
compare intermediate states of language competence to both the beginning (no
competence) and end (native competence) states of language learning. And by
confining the study to learners who have had qualitatively and quantitatively similar
exposure to the language, one avoids the problems of interpretation introduced by
variable linguistic experiences across subject groups. However, even under ideal
circumstances, uncertainties often exist (for reasons discussed above) concerning
the relationship between the BOLD signal and the transient events underlying lan-
guage comprehension.
In principle, ERPs provide the temporal resolution and sensitivity needed to
reveal how (and potentially where) the L2 is processed, how similar this processing
is to L1 processing, and how L2 processing changes as a function of increasing L2 Page 287 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
experience. In a series of ERP experiments, we have tracked changes in brain activity
that are correlated with increasing exposure to the L2 in adult L2 learners. We
describe here two experiments, one involving L2 word learning (McLaughlin, Oster-
hout, & Kim, in preparation), and a second involving sentences (Osterhout et al., in
preparation). In both studies, English-speaking novice French learners were studied
longitudinally as they progressed through their first year of classroom French instruc-
tion. Learners were tested three times: near the beginning, in the middle, and near
the end of the instructional period. This design allowed us to track changes in brain
activity within a single group of language learners, and ensured that the learners
were relatively homogeneous in their initial lack of L2 proficiency and in their
subsequent L2 experience.
McLaughlin et al. examined word learning by measuring learning-related
changes in N400 amplitude, which is sensitive to both lexical status (i.e., whether
or not a letter string is a word) and word meaning (Bentin, 1987; Kutas & Hillyard,
1980). For native speakers, N400 amplitude is largest for pronounceable, orthograph-
ically legal nonwords (pseudowords, e.g., flirth), intermediate for words preceded
by a semantically unrelated context, and smallest for words preceded by a seman-
tically related context. Our goal was to determine how much L2 exposure is needed
before the language-learners’ brain activity elicited by L2 words and word-like
stimuli resembles that of a native speaker.
McLaughlin et al. asked adult French learners to make word/nonword judgments
to a sequence of French words and pronounceable nonwords. They found that after
only 14 hr (roughly two weeks) of L2 instruction, the nonwords elicited a robustly
larger-amplitude N400 than did words (Figure 14.6). This was true even though
learners showed random behavior (d-prime = 0) when making explicit word/nonword
judgments about these stimuli. Furthermore, the correlation between hours of instruc-
tion and the word/nonword N400 difference was very robust (r = .74), suggesting
that the N400 difference was approximately a linear function of the amount of L2
exposure. Effects of word meaning, manifested as smaller-amplitude N400s to words
preceded by related than by unrelated words, were observed after approximately 62
hr of instruction. By the end of the instructional period (after approximately 138
hr), the amplitude of the word/nonword differences approximated that typically
observed in native speakers, even though learners’ conscious lexicality judgments
remained very poor (d-prime < 1). Importantly, no N400 differences were observed
for a group of subjects who had received no French instruction.
Thus, even for adult learners, the brain response to L2 words and word-like
nonwords changes dramatically with very little L2 exposure. Furthermore, these
changes are not random, but instead, almost immediately begin to approximate the
responses seen in native speakers to analogous stimuli, and occur even while the
learner behaves randomly when making consciously lexicality judgments to these
stimuli. These results seem to indicate that the brain response to words and word-
like strings in the L2 changes dramatically with remarkably little L2 exposure.
A second experiment was motivated by the observation, discussed above, that
syntactic and semantic anomalies embedded within sentences elicit distinct ERP
effects (the P600 and N400 effects, respectively). With respect to adult L2 learning,
several questions immediately come to mind: How much experience with a language
AU: Not
included in
ref list.
AU: Not
included in
ref list.
AU: Not
included in
ref list. Page 288 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
is needed for the learner’s brain to distinguish between well-formed and ill-formed
sentences? How much experience is needed for the learner’s brain to distinguish
between different types of linguistic anomalies (e.g., syntactic and semantic)? And
how much experience is needed for the learner’s brain to respond to these anomalies
like the brain of a native speaker of the L2? To answer these questions, we presented
well-formed French sentences and French sentences containing either a semantic or
a syntactic anomaly, as in the below examples:
(3) Sept plus cinq\*livre font douze. (semantic condition)
(4) Tu adores\*adorez le francais. (verb conjunction condition)
Figure 14.6. ERPs (recorded over the vertex) for No Instruction (left panel) and French
Instruction (right panel) subjects, recorded at the three successive testing sessions. ERPs are
plotted for three types of target strings: target words that were semantically related to the
prime word (solid line), target words that were not semantically related to the prime word
(small dashes), and target nonwords (large dashes).
200 400 600
Related words (target)
Non-related words (target)
Non-words (target)
Session 1
Session 2
Session 3
No French Instruction French Leamers
N400 Page 289 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
(5) Tu manges des hamburgers\*hamburger pour diner. (article–noun agree-
ment condition)
In (3), the noun livre is semantically anomalous. In (4), the verb adorez is
conjugated incorrectly, given the preceding sentence fragment. In (5), the noun
hamburger disagrees with the syntactic number of the plural article. Thus, the
anomalous version of (3) is semantically anomalous, whereas the anomalous versions
of (4) and (5) are syntactically anomalous. Importantly, the two syntactic rules differ
in terms of their relative prevalence in the L1 and L2. Verb conjugation in French
is quite similar to that in English; however, article–noun agreement is exceedingly
common in French but is much less so in English.
Subjects were native French speakers and a group of novice, adult French
learners. Learners were tested after approximately 1 month, 4 months, and 8 months
of French instruction. As expected, native French speakers showed an N400 effect
to the semantically anomalous words, and very large P600 effects to the two types
of syntactic anomalies. The French learners, as is often the case, showed striking
individual differences, both in a behavioral “sentence acceptability judgment” task
and in the pattern of ERPs elicited by the anomalous stimuli. We then segregated
the learners into upper (“fast learners”) and lower (“slow learners”) halves, based
on their performance on the sentence-acceptability judgment task, and averaged the
ERPs separately for each group. Results for the “fast learner” group will be discussed
here. At each testing session, including the initial session that occurred after just 1
month of instruction, semantically anomalous words elicited a robust N400 effect,
and this effect changed minimally with increasing instruction (Figure 14.7). The
finding of real interest pertained to the two syntactic conditions. We will describe
the verb conjugation condition first (Figure 14.8). After just 1 month of instruction,
the learners’ brains discriminated between the syntactically well-formed and ill-
formed sentences. However, rather than eliciting the P600 effect (as we saw in native
French speakers), the syntactically anomalous words elicited an N400-like effect.
By 4 months, the N400 effect had largely disappeared and was replaced by a P600-
like positivity. By 8 months, this P600 effect increased in amplitude and the N400
effect was entirely absent.
One interpretation of these results is that the learners initially learned about words
but not rules. That is, after one month of French instruction they recognized, for
example, that the word adorez does not fit well in the context “Tu … ” but had yet
to grammaticalize this knowledge (and hence elicited an N400 effect). With a bit
more exposure, this knowledge became codified as a syntactic rule (and hence elicited
a P600 effect). Regardless of the validity of this speculation, this experiment, much
like the McLaughlin et al. experiment, demonstrates the dramatic changes in the brain
responses to tokens of the L2 that occur with very little L2 exposure. As in the
McLaughlin et al. study, less than a year of L2 exposure was sufficient to produce
brain responses that were qualitatively similar to native-speaker brain responses.8
These findings seem to argue in favor of a great deal of brain plasticity, even
with minimal adult L2 instruction. However, the results in the article–noun agree-
ment condition clearly demonstrated limits to this plasticity: Learners’ judgments
concerning the grammaticality of these sentences were uniformly poor (even after
AU: Not
included in
ref. list. Page 290 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
a year of L2 instruction), and no differences were observed in the ERPs to critical
words in the grammatical and ungrammatical sentences (Figure 14.9). This was true
even though article–noun number agreement is seemingly a simple rule, and even
though both rules were encountered at roughly the same (early) point in the instruc-
tional period. Informal discussions with the French instructors indicated that learners
often have more trouble learning the agreement rule than the conjugation rule.
Although the proper interpretation of this set of results is uncertain at the moment,
one possibility is that the rate of L2 learning is determined in part by L1–L2
similarity. Those aspects of the L2 that are highly similar to the L1 will be learned
very rapidly; those aspects that are sufficiently dissimilar will be learned very slowly.
Dissimilarities might include the presence of a rule in one language and its absence
in the other language, differences in how the rule is expressed across languages, or
differences in the prevalence or salience of that rule across languages. This proposal
is not new, as influences of L1–L2 similarity on L2 learning have been documented
for many years (Gass & Selinker, 1992; Odlin, 1989, Ringbom, 1987).
In summary, these findings seem to demonstrate that dramatic changes in brain
activity occur during the earliest stages of adult L2 learning. Furthermore, these
changes are sometimes, but not always, accompanied by increasing accuracy in
related behavioral judgments. Finally, the qualitative nature of these changes might
reveal important details of what has been learned. Importantly, however, mere expo-
sure to the L2 is not enough to guarantee increasing sensitivity to every aspect of
the language; these rapid changes in brain activity might depend on similarities
across the learner’s L1 and L2.
Figure 14.7. ERPs (recorded over the vertex) to critical words in the well-formed (solid
line) and semantically anomalous (dashed line) conditions, collapsed over the three testing
5 µv
200 400 600 800 1000
Semantically anomalous Page 291 Tuesday, March 23, 2004 4:49 PM
Figure 14.8. ERPs (recorded over the vertex) to critical words in the well-formed (solid line) and verb conjugation anomaly (dashed line)
conditions, plotted separately for each of the three testing sessions.
Session 1
5 µv
200 400 600 800 1000
Verb conjugation anomaly
Session 2
Session 3
P600 Page 292 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
As is the case with any method, ERPs have their limitations. The most fundamental
limitation is one we noted earlier in this chapter: It is relatively easy to identify the
stimulus manipulations (that is, the antecedent conditions) that produce or modulate
some ERP effect, but it is often very difficult to identify the precise cognitive events
underlying the effect.9 Misattributing a particular function to a particular ERP effect
can have serious consequences, particularly with respect to developing and testing
theories of language processing. To illustrate this point, we will discuss two example
cases in which particular ERP effects have been tied to specific cognitive processes.
In both examples, we present evidence that at the very least raises some questions
about the general veracity of these claims. The goal of this exercise is to demonstrate
the difficulty of ascertaining with any confidence the precise process made manifest
by some ERP effect. We are not claiming that these claims are universally wrong,
nor are we disparaging the science that motivates them.
The first example concerns the distinction between content (or “open-class”)
and function (or “closed-class”) words (see Hoen and Dominey, chapter 16, this
volume). Content words (nouns, verbs, adjectives, etc.) play a largely referential role
in language, whereas function words (articles, prepositions, auxiliary verbs, etc.)
play a primarily syntactic role. Neville and colleagues (e.g., Neville, Mills, &
Figure 14.9. ERPs (recorded over the vertex) to critical words in the well-formed (solid
line) and article–noun agreement anomaly (dashed lines) conditions, collapsing over the three
testing sessions.
5 µv
200 400 600 800 1000
Article-noun agreement anomaly Page 293 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
Lawson, 1992) have argued that content words elicit a negative-going wave that
peaks at about 400 ms (N400), whereas function words elicit a negativity that peaks
at about 280 ms (N280). The N400 component is largest over the posterior parts of
the scalp, whereas the N280 is largest over anterior portions of the left hemisphere.
These areas of maximum amplitude are interesting because they correspond to the
lesion sites that putatively produce problems in processing the meaning and form
of sentences, respectively. Such evidence has led Neville and colleagues to conclude
that the N400 and N280 reflect specifically the semantic and syntactic functions of
these two word classes, respectively.
The problem with this claim is that many variables are confounded with the
content/function distinction. Most notable in this regard is the confounding between
word class, on the one hand, and word length and normative word frequency, on the
other. Function words tend to be shorter and more frequent than content words. It
could be, then, that the ERP differences between content and function words are
due mostly to the physical properties of length and frequency than to their abstract
linguistic functions. To investigate this possibility, we asked subjects to read a short
essay for comprehension (Osterhout, Allen, & McLaughlin, 2002). We then averaged
ERPs in two ways: as a function of word class (content and function) and as a
function of word length. When we averaged the ERPs as a function of word class,
we replicated the result reported by Neville and colleagues. Function words elicited
a negative component at about 300 ms that was largest over anterior portions of the
left hemisphere, whereas content words elicited a posterior-maximal N400 compo-
nent (Figure 14.10a). However, when we averaged the ERPs as a function of word
length, we found that the variation in amplitude and latency could be accounted for
almost entirely by the single variable of word length (Figure 14.10b). This robust
near-linear function is much more consistent with a “word-length” model than a
“word-class” model; the word-length model predicts a continuous distribution,
whereas the word-class model predicts a bimodal one.
AU/PE: Capi-
tal A and
B in the
caption and
AU/PE: Capi-
tal A and
B in the
caption and
200 400 600 800
Function words
Content words
N300 Page 294 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
Our second example concerns the observation, noted above, that certain types
of syntactic (but not semantic or pragmatic) anomalies elicit a left anterior negativity
(LAN) in addition to eliciting a P600 effect. Friederici and colleagues (Hahne &
Friederici, 1999; Friederci, 1995, 2002) have proposed a two-stage functional model
for these ERP effects. They claim that the LAN effect reflects a fast, automatic
syntactic analyzer, and that the P600 effect reflects attempts at syntactic reanalysis.10
Underlying these claims are three more basic claims: first, that the LAN effect is
reliably elicited by violations of syntactic rules; second, that the LAN effect can be
associated with a neural source in or near Broca’s area (Friederici, Hahne, & von
Cramon, 1998; Friederici, von Cramon, & Kotz, 1999; Friederici, Wang, Herrmann,
Maess, & Oertel, 2000); and, third, that (on a single trial and within a single subject)
the ERP response to syntactic rule violations is biphasic, involving an initial detection
of the error (reflected in the LAN) and then an attempt to fix it (reflected in the P600).
Each of these claims is open to debate. First, a critical step in identifying the
cognitive processes made manifest by some ERP effect is to identify the antecedent
conditions that elicit or modulate it. However, the antecedent conditions that elicit
the LAN effect are not clear. Although LAN effects are often reported in the response
to a syntactic anomaly, there are a significant number of reports in which they are
not reported (e.g., Ainsworth- Darnell, Shulman, & Boland,1998; Allen et al., in
press; Hagoort, Brown, & Groothusen, 1994; Kuperberg, Holcomb, et al., 2003;
McKinnon & Osterhout, 1996; Osterhout, Bersick, & McLaughlin, 1997; Osterhout,
McLaughlin, & Inoue, 2002; Osterhout & Mobley, 1995; Takazawa et al., 2002).
Figure 14.10. (A) ERPs (recorded over anterior midline site Fz) averaged as a function of
word class. Adapted from “Words in the brain: Lexical determinants of word-induced brain
activity,” by L. Osterhout, M. Allen, and J. McLaughlin, 2002, Journal of Neurolinguistics,
15, 171–187. (B) ERPs averaged as a function of word length. Adapted from “Words in the
brain: Lexical determinants of word-induced brain activity,” by L. Osterhout, M. Allen, and
J. McLaughlin, 2002, Journal of Neurolinguistics, 15, 171–187.
200 400 600 800
length 2
length 4
length 6
AU/PE: Low-
ercase a
and b used
in text. AU/PE: Low-
ercase a
and b used
in text.
AU: 1993 in
ref. list. Page 295 Tuesday, March 23, 2004 4:49 PM
Second, the distribution of the LAN effect is somewhat variable and is not infre-
quently reported to have a bilateral (e.g., Friederici et al., 1999) or even a right-
hemisphere-maximal distribution (e.g., Osterhout & Nicol, 1999, Exp. 2). This
variability raises questions about the claim that the neural generators of the effect
lie in or near Broca’s area.
Third, even given the apparent existence of a LAN effect in the brain response
to syntactic rule violations, it is not certain that individual subjects in individual
trials show a biphasic response (LAN followed by P600) to these anomalies. This
is because ERPs in the relevant reports were averaged (as is usually done) over
subjects and trials. Thus, it is conceivable that the syntactic anomalies in Friederici’s
experiments elicited a negative-going effect in some subjects and a P600-like
response in others. Such a finding would be problematic for Friederici’s multistage
model. A result much like this has been reported in one recent study (Inoue &
Osterhout, in preparation). Inoue and Osterhout asked Japanese speakers to read
sentences such as the following (written here in an English gloss):
(6a) Taro-NOM Hanako-DAT textbook-ACC to-buy said.
“Taro told Hanako to buy a textbook.
(6b) *Taro-NOM Hanako-ACC textbook-ACC to-buy said.
*“Taro told ??? to buy a textbook/Hanako.
In Japanese, the verb typically appears at the end of the clause, and the nouns
are attached to case markers specifying their grammatical and thematic roles. In
both (6a) and (6b), the subject of the sentence is followed by two nouns. In (6a),
these nouns are appropriately marked for dative and accusative roles, respectively.
However, in (6b), both nouns are marked for the accusative role. If readers assume
that both nouns are attached to the main verb, the second noun should be perceived
to be anomalous.11
Therefore, ERPs to the second noun in sentences like (6a) and (6b) were of
interest. When we averaged ERPs over all subjects, the response to the syntactically
anomalous words seemed to be biphasic, involving both an anterior negativity
between 300 and 500 ms, and then a large posterior P600 effect (Figure 14.11a).
However, when subjects were divided into two groups based on the magnitude of
the anterior negativity effect, ERPs averaged separately for these two groups revealed
compelling individual differences: Those subjects who responded with a large neg-
ativity to the violations tended to show a very small P600 effect, whereas those that
responded with a large P600 effect showed no evidence of a negativity (Figures
14.11b and 14.11c). Furthermore, the “anterior” distribution of the negativity largely
disappeared when averages were formed over the two groups; in fact, the timing
and distribution of the negativity elicited by these case anomalies were similar in
this group of subjects to the timing and distribution of N400 elicited by semantic
anomalies. Apparently, the anterior distribution was (at least to a degree) an artifact
caused by component overlap (i.e., the posteriorly distributed P600 effect reduced
the size of the negativity over posterior sites, but not over anterior sites).
Obviously, it is far from clear how this finding relates to prior reports of a “LAN
followed by P600” response to different types of syntactic anomalies in English and
AU: Not
included in
ref. list.
AU/PE: Capi-
tal A, B,
and C in
caption and
AU/PE: Capi-
tal A, B,
and C in
the caption
and figure. Page 296 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
German. The relevant point, however, is that sometimes the grand average does not
accurately reflect what occurs on a single trial or within a single subject. Unfortu-
nately, if the sample is comprised of two or more populations, each of which responds
differently to some event, this fact can be obscured by averaging over everyone.12
This, in turn, can lead to erroneous conclusions about the distribution of an effect
across the scalp, the ordering of events during a particular process, and the precise
cognitive processes underlying some effect.
Perhaps it will turn out that the LAN effect is reliably elicited by syntactic
anomalies, that it has a consistent scalp distribution, and that it is not an artifact of
averaging over subjects. Is such evidence sufficient to justify the conclusion that the
LAN effect directly manifests a fast, automatic syntactic analyzer? Unfortunately,
no. This is because (as we noted above) the processes underlying the LAN effect
(or the P600) might be correlated with, but indeterminately removed from, the
syntactic processes themselves. One is still limited in the types of inferences that
are licensed by the evidence; for example, by localizing the source or the LAN effect
or the P600, one has not definitively isolated the source of syntactic processing. In
all likelihood, definitive conclusions concerning the cognitive processes underlying
these ERP effects will require converging evidence from other methods of investi-
Friederici and colleagues provide another type of evidence that the LAN effect
reflects the detection of a syntactic anomaly, whereas the P600 reflects syntactic
Figure 14.11. (A) ERPs averaged over 20 subjects (recorded over an anterior and a posterior
site) elicited by Japanese sentences of the form “Noun-Nominative Noun-Dative Noun-
Accusative” (solid line) and “Noun-Nominative Noun-Accusative Noun-Accusative” (dashed
line). Readers are predicted to encounter processing problems at the second of two consecutive
nouns marked for accusative case. ERPs shown are those elicited to the second post-subject
noun (Noun-Accusative) in each sentence type. (B) ERPs averaged over 10 subjects who
showed the smallest-magnitude Anterior Negativity. (C) ERPs averaged over 10 subjects who
showed the largest-magnitude Anterior Negativity.
Anterior Negativity Negativity
Indirect Object Direct Object
Direct Object Direct Object
5 µv
200 400 600 800 1000
AU/PE: Low-
ercase a,
b, and c
used in text.
AU/PE: Low-
ercase a,
b, and c
used in text.
AU/PE: Low-
ercase a,
b, and c
used in text. Page 297 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
reanalysis. They have shown that although the P600 is sensitive to stimulus manip-
ulations (e.g., manipulating the proportion of anomalies within a list) and task
manipulations (e.g., monitoring for a syntactic vs. a semantic problem), the LAN
effect is not (Hahne & Friederici, 1999, 2002). Such a result is expected if the LAN
reflects the actions of a fast, automatic syntactic analyzer, whereas the P600 does
not. However, these demonstrations suffer from a potentially serious materials flaw.
In both of their experiments, Hahne and Friederici presented their sentences as
continuous natural speech. In such a paradigm, ERPs to several successive words
will commingle. If these words are different across conditions, confounds are likely.
In the Hahne and Friederici experiments, sentences in the control and syntactically
anomalous conditions were comprised of sequences such as (1) Das Brot wurde
gegessen (The bread was eaten) and (2) Das Brot wurde im gegessen (The bread
was in-the eaten), respectively. The critical word was the verb gegessen. This com-
parison is problematic. For example, wurde and im might be different in phonological
length and/or normative frequency, and thereby produce differences in ERPs that
extend into the critical word epoch. Furthermore, it is likely that ERPs to Brot and
wurde overlapped with the ERPs to gegessen in (1), while ERPs to wurde and im
overlapped with the response to gegessen in (2). Nouns like Brot and high-frequency
function words like wurde and im elicit quite different ERPs (e.g., Neville et al.,
1992; Osterhout et al., 2002). Hence, one cannot assume that the reported differences
in the ERPs to gegessen are due to the ungrammaticality of (2). Importantly, stimulus
confounds of this sort would not be influenced by task and stimulus manipulations.
This potential problem is not ameliorated by the use of a poststimulus baseline to
equate the well-formed and ill-formed conditions (as was done by Hahne and
Friederici); whatever preexisting differences are present continue to resolve them-
selves and continue to contaminate ERPs to the critical words, regardless of the
choice in baseline. Although the existence or nonexistence of these confounds could
probably be ascertained by examining the prestimulus portion of the epoch elicited
by the critical word, Hahne and Friederici go counter to convention and do not plot
this portion of the waveform.
It is undeniably true that the more we know about the cognitive events underlying
some language-sensitive ERP effect, the more useful that ERP effect becomes. It is
also true that erroneous hypotheses of this sort can have serious and unfortunate
consequences, especially with respect to theory-building. We would like to be clear
about what we are claiming: The hypotheses discussed in this section are useful.
Even so, it is important to maintain a cautious and realistic attitude concerning the
difficulties in attempting to validate such hypotheses.
Fortunately, these realities do not limit in any profound way the utility of ERPs
for studying sentence processing. All that is needed to make progress is the discovery
of ERP effects that co-vary systematically with principled stimulus manipulations
(cf. Osterhout & Holcomb, 1995). With that in hand, one can use the ERP data to
contrast functional claims about sentence processing even if one cannot make clear
functional claims about the ERP effects themselves. The preceding two decades of
work has provided convincing evidence of such effects and of their utility as tools
for studying real-time sentence comprehension (see also chapter 13 by van Berkum,
chapter 15 by Barber et al., and chapter 16 by Hoen and Dominey in this volume). Page 298 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
In our review, we have attempted to illustrate the advantages of ERPs for studying
real-time language comprehension, both in native speakers and in adult language
learners. We recognize, of course, that ERPs (like all methods of investigation)
imperfectly reflect the cognitive and neural processes underlying human language.
Because of this, it is wise to consider possible sources of converging evidence
whenever possible.
Nonetheless, sentence comprehension (like all aspects of language comprehen-
sion) involves complex events that occur with great speed. A satisfying theory of
sentence comprehension will explain how these events occur over time, in real time,
as a person is trying to understand a sentence. The same is true for language learning,
although in that case the theory must track changes that occur over two vastly
different time scales: the time it takes a person to understand a word or sentence
and the time it takes a person to learn the facts of the foreign language. A truly
compelling theory will link these events to the brain, relying on direct measurement
of brain activity. The primary (and perhaps unique) advantage of ERPs is that they
provide us with a means for observing reasonably direct manifestations of some of
the transient brain events that make up the core of language processing. ERPs also
allow us to determine the “response selectivities” of these brain events through
careful manipulation of linguistic stimuli. Given recent methodological advances, it
appears that we might now be able to more compellingly relate these brain events
to their underlying neural sources.
A cynically minded reader might argue that some of the novel insights described
above are a bit too novel to be entirely believable. Surely we know that syntax is
always first, that critical periods for language learning exist, and that fMRI is better
suited than ERPs for localizing neural activity in the brain. We do not mean to
espouse dangerously or ludicrously radical ideas. We do mean to suggest, however,
that a particular method should be evaluated on the basis of its a priori suitability,
its sensitivity to the phenomena of interest, and its ability to generate reproducible
and theoretically interesting results. ERPs get high marks along all of these dimen-
sions. ERPs (or any other method) should not be devalued or viewed skeptically
simply because the data generated by the method seemingly contradict conventional
beliefs. On the contrary, such methods should be valued more highly, as they provide
the primary means for advancing our theoretical understanding.
Ainsworth-Darnell, K., Shulman, R., & Boland, J. (1998). Dissociating brain responses to
syntactic and semantic anomalies: Evidence from event-related brain potentials. Jour-
nal of Memory and Language, 38, 112–130.
Allen, M. D., Badecker, W., & Osterhout, L. (in press). Morphological analysis during
sentence processing. Language and Cognitive Processes. Page 299 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
Anderer, P., Pascual-Marqui, R. D., Smlitsch, H. V., & Saletu, B. Differential effects of normal
aging on sources of standard N1, target N1, and target P300 auditory event-related
brain potentials revealed by low resolution electromagnetic tomography (LORETA).
Electroencephalography and Clinical Neurophysiology, 108, 160–174.
Angrilli, A., Penolazzi, B., Vespignani, F., De Vincenzi, M., Job, R., Ciccarelli, L., Palomba,
D., & Stegagno, L. (2002). Cortical brain responses to semantic incongruity and
syntactic violation in Italian language: An event-related potential study. Neuroscience
Letters, 322, 5–8.
Bates, E., & Dick, F. (2000). Beyond phrenology: Brain and language in the next millennium.
Brain and Language, 71, 18–21.
Bavelier, D., Corina, D., Jezzard, P. et al. (1997). Sentence reading: A functional MRI study
at 4 tesla. Journal of Cognitive Neuroscience, 9, 664–686.
Bentin, S. (1987). Event-related potentials, semantic processes, and expectancy factors in
word recognition. Brain and Language, 31, 308–327
Bever, T. G., & Townsend, D. J. (2001). Some sentences on our consciousness of sentences.
In E. Dupoux (Ed.), Language, brain, and cognitive development: Essays in honor
of Jacques Mehler. Cambridge, MA: MIT Press.
Binder, J., & Price, C. J. (2001). Functional neuroimaging of language. In Cabeza & A.
Kingstone (Ed.), Handbook of functional neuroimaging of cognition. Cambridge, MA:
MIT Press.
Blumstein, S., & Milberg, W. (2000). Language deficit in Broca’s and Wernicke’s aphasia:
A singular impairment. In Y. Grodzinsky, L. Shapro, & D. Swinney (Eds.), Language
and the brain: Representation and processing. San Diego, CA: Academic Press.
Bock, J. K., & Kroch, A. S. (1989). The isolability of syntactic processing, In G. N. Carlson
& M. K. Tanenhaus (Eds.), Linguistic structure in language processing. Boston:
Kluwer Academic.
Burock, M. A., Buckner, R. L., Woldorff, M. G., Rosen, B. R., & Dale, A. M. (1998).
Randomized event-related experimental designs allow for extremely rapid presenta-
tion rates using functional MRI. NeuroReport, 9, 3735–3739.
Caplan, D., Hildebrandt, N., & Makris, N. (1996). Location of lesions in stroke patients with
deficits in syntactic processing in sentence comprehension. Brain, 119, 933–949.
Caramazza, A., & Zurif, E. B. (1976). Dissociation of algorithmic and heuristic processes in
language comprehension: Evidence from aphasia. Brain and Language, 3, 572–582.
Chee, M. W., Caplan, D., Soon, C. S., Sriram, N., Tan, E. W., Thiel, T., et al. (1999a).
Processing of visually presented sentences in Mandarin and English studied with
fMRI. Neuron, 23, 127–137.
Chee, M. W., Hon, N., Lee, H. L., & Soon, C. S. (2001). Relative language proficiency
modulates BOLD signal change when bilinguals perform semantic judgments. Neu-
roimage, 13, 1155–1163.
Chee, M. W., Tan, E. W., & Thiel, T. (1999b). Mandarin and English single word processing
studied with functional magnetic resonance imaging. The Journal of Neuroscience,
19, 3050–3056.
Chomsky, N. (1986). Knowledge of language. New York: Praeger.
Dale A. M, Bunker R. L. (1997). Selective averaging of individual trials using fMRI. Human
Brain Mapping, 5, 329–340.
Dale, A. M., Liu, A. K., Fischl, B. R., Buckner, R. L., Belliveau, J. W., Lewine, J. D., et al.
(2000). Dynamic statistical parameteric mapping: Combining fMRI and MEG for
high-resolution imaging of cortical activity. Neuron, 26, 55–67.
AU: Pl. pro-
vide all
AU: Pl. pro-
vide editors
AU: 2002 in
text citation. Page 300 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
Dehaene, S., Dupoux, E., Mehler, J., Cohen, L., Paulesu, E., Perani, D., et al. (1997).
Anatomical variability in the cortical representation of first and second language.
Neuroreport, 8, 3809–3815.
Donchin, E., & Coles, M. G. H. (1988). Is the P300 component a manifestation of context
updating? Behavioral and Brain Sciences, 11, 343–427.
Dronkers, N. F. (2000). The pursuit of brain-language relationships. Brain and Language, 71,
Ellis, A. W., & Lambon Ralph, M. A. (2000). Age of acquisition effects in adult lexical
processing reflect loss of plasticity in maturing systems: Insights from connectionist
networks. Journal of Experimental Psychology: Learning, Memory, Cognition, 26,
Elman, J. L. (1990). Representation and structure in connectionist models. In G. T. M. Altmann
(Ed.), Cognitive models of speech processing. Cambridge, MA: MIT Press.
Ferreira, F., & Clifton, C. (1986). The independence of syntactic processing. Journal of
Memory and Language, 25, 348–368.
Fodor, J. A. (1983). Modularity of mind. Cambridge, MA: MIT Press.
Fodor, J. D., & Ferreira, F. (1998). Reanalysis in sentence processing. Boston: Kluwer
Forster, K. (1979). Levels of processing and the structure of the language processor. In W.
Cooper & E. C. T. Walker (Eds.), Sentence processing. Hillsdale, NJ: Erlbaum.
Frazier, L. (1989). Against lexical generation of syntax. In W. Marslen-Wilson (Ed.), Lexical
representation and process. Cambridge, MA: MIT Press.
Frazier, L., & Clifton, C., Jr. (1996). Construal. Cambridge, MA: MIT Press.
Friederici, A. D. (1995). The time course of syntactic activation during language processing:
A model based on neuropsychological and neurophysiological data. Brain and Lan-
guage, 50, 259–284.
Friederici, A. D. (2002). Towards a neural basis of auditory sentence processing. Trend s in
Cognitive Sciences, 6, 78–84.
Friederici, A. D., Hahne, A., & Saddy, D. (2002). Distinct neurophysiological patterns reflect-
ing aspects of syntactic complexity and syntactic repair. Journal of Psycholinguistic
Research, 31, 45–63.
Friederici, A. D., Hahne, A., & von Cramon, D. Y. (1998). First-pass versus second-pass
parsing processes in a Wernicke’s and a Broca’s aphasic: Electro-physiological evi-
dence for a double dissociation. Brain and Language, 62, 311–341.
Friederici, A. D., Mecklinger, A., & Hahne, A. (1996). The temporal structure of syntactic
parsing: Early and late event-related brain potential effects elicited by syntactic
anomalies. Journal of Experimental psychology: Learning, Memory, and Cognition,
22, 1219–1248.
Friederici, A. D., Ruschemeyer, S., Hahne, A., & Fiebach, C. J. (2003). The role of left
inferior frontal and superior temporal cortex in sentence comprehension: Localizing
syntactic and semantic processes. Cerebral Cortex, 13, 170–177.
Friederici, A. D., von Cramon, D. Y., & Kotz, S. A. (1999). Language related brain potentials
in patients with cortical and subcortical left hemisphere lesions. Brain, 122,
Friederici, A. D., Wang, Y., Herrmann, C. S., Maess, B., & Oertel, U. (2000). Localization
of early syntactic processes in frontal and temporal cortical areas: A magnetoenceph-
alographic study. Human Brain Mapping, 11, 1–11.
Garnsey, S. M., Pearlmutter, N. J., Myers, E., & Lotocky, M. A. (1997). The contributions
of verb bias and plausibility to the comprehension of temporarily ambiguous sen-
tences. Journal of Memory and Language, 37, 58–93.
AU: Not cited
in text.
AU: Not cited
in text.
AU: Not cited
in text.
AU: Not cited
in text. Page 301 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
Gass, S., & Selinker, L. (1992). Language Transfer in Language Learning. Amsterdam: John
Haan, H., Streb, J., Bien, S., & Roesler, F. (2000). Individual cortical current density recon-
structions of the semantic N400 effect: Using a generalized minimum norm model
with different constraints (L1 and L2 norm). Human Brain Mapping, 11, 178–192.
Hagoort, P., Brown, C., & Groothusen, J. (1993). The syntactic positive shift (SPS) as an
ERP measure of syntactic processing. Language and Cognitive Processes, 8, 439–483.
Hahne, A., & Fiederici., A. D. (1999). Electrophysiological evidence for two steps in syntactic
analysis: Early automatic and late controlled processes. Journal of Cognitive Neuro-
science, 11, 193–204.
Hahne, A., & Friederici., A. D. (2001). Processing a second language: Late learners’ com-
prehension mechanisms as revealed by event-related brain potentials. Bilingualism:
Language and Cognition, 4, 123–141.
Hahne, A., & Friederici, A. D. (2002). Differential task effects on semantic and syntactic
processes as revealed by ERPs. Cognitive Brain Research, 13, 339–356.
Halgren, E., Dhond, R. P., Christensen, N., van Petten, C., Marinkovic, K., Lewine, J., &
Dale, A. M. (2002). N400-like magnetoencephalography responses modulated by
semantic context, word frequency, and lexical class in sentences. NeuroImage, 17,
Halsband, U., Krause, B. J., Sipila, H., Teras, M., & Laihinen, A. (2002). PET studies on the
memory processing of word pairs in bilingual Finnish-English subjects. Behavioral
Brain Research, 132, 47–57.
Hämäläinen, M. S., & Sarvas, J. (1989). Realistic conductivity geometry model of the human
head for interpretation of neuromagnetic data. IEEE Transactions on Biomedical
Engineering, 36, 165–171.
Hasegawa, M., Carpenter, P. A., & Just, M. A. (2002). An fMRI study of bilingual sentence
comprehension and workload. NeuroImage, 15, 647–660.
Henderson, C. J., Butler, S. R., & Glass, A. (1975). The localization of equivalent dipoles of
EEG sources by the application of electrical field theory. Electroencephalography
and Clinical Neurophysiology, 39, 117–130.
Hernandez, A. E., Martinez, A., & Kohnert, K. (2000). In search of the language switch: An
fMRI study of picture naming in Spanish-English bilinguals. Brain and Language,
73, 421–431.
Illes, J., Francis, W. S., Desmond, J. E., Gabrieli, J. D., Glover, G. H., Poldrack, R., Lee, C.
J., & Wagner, A. D. (1999). Convergent cortical representation of semantic processing
in bilinguals. Brain and Language, 70, 347–363.
Jackendoff, R. (2000). Fodorian and representational modularity. In Y. Grodzinsky, L. Shapiro,
& D. Swinney (Eds.), Language and the brain: Representation and processing. San
Diego, CA: Academic Press.
Jezzard, P., Matthews, P. M., & Smith, S. (2001). Functional magnetic resonance imaging:
Methods for Neuroscience. Oxford: Oxford University Press.
Johnson, J. S. (1992). Critical period effects in second language acquisition: The effect of
written versus auditory materials on the assessment of grammatical competence.
Language Learning, 42, 217–248.
Johnson, J. S., & Newport, E. L. (1989). Critical period effects in second language learning:
The influence of maturational state on the acquisition of English as a second language.
Cognitive Psychology, 21, 60–99.
Kang, A. M., Constable, R. T., Gore, J. C., & Avrutin, S. (1999). An event-related fMRI study
of implicit phrase-level syntactic and semantic processing. Neuroimage, 10, 555–561. Page 302 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
Kavanaugh, R. N., Darcey, T. M., Lehmann, D., & Fender, D. H. (1978). Evaluation of methods
for three dimensional localization of electrical sources in the human brain. IEEE
Transactions in Biomedical Engineering, 25, 421–429.
Kim, K. H. S., Relkin, N. R., Lee, K-M., & Hirsch, J. (1997). Distinct cortical areas associated
with native and second languages. Nature, 388, 171–174.
Klein, D., Milner, B., Zatorre, R. J., Zhao, V., & Nikelski, J. (1999). Cerebral organization
in bilinguals: A PET study of Chinese-English verb generation. NeuroReport, 10,
Kolk, H. H. J., Chwilla, D. J., van Herten, M., & Oor, P. J. W. (2003). Structure and limited
capacity in verbal working memory: A study with event-related potentials. Brain and
Language, 85, 1–36.
Kounios, J., Smith, R. W., Yang, W., Bachman, P., & D’Esposito, M. (2001). Cognitive
association formation in human memory revealed by spatiotemporal brain imaging.
Neuron, 29, 297–306.
Kuhl, K. P. (2000). Language, mind, brain: Experience alters perception. In S. M. Gazzaniga
& E. Bizzi (Eds.), The new cognitive neurosciences. Cambridge, MA: MIT Press.
Kuperberg, G. R., Holcomb, P. J., Sitnikva, T., Greve, D., Dale, A. M., & Caplan, D. (2003).
Distinct patterns of neural modulation during the processing of conceptual and syn-
tactic anomalies. Journal of Cognitive Neuroscience, 15, 272–293.
Kuperberg, G. R., McGuire, P. K., Bullmore, E. T., Brammer, M. J., Rabe-Hesketh, S., Wright,
I. C., et al. (2000). Common and distinct neural substrates of pragmatic, semantic,
and syntactic processing of spoken sentences. Journal of Cognitive Neuroscience, 12,
Kuperberg, G. R., Sitnikova, T., Caplan, D., & Holcomb, P. J. (2003). Electrophysiological
distinctions in processing conceptual relationships within simple sentences. Cognitive
Brain Research, 17, 117–129.
Kutas, M., & Hillyard, S. A. (1980). Reading senseless sentences: Brain potentials reflect
semantic incongruity. Science, 207, 203–205.
Kutas, M., & Hillyard, S. A. (1984). Brain potentials during reading reflect word expectancy
and semantic association. Nature, 307, 161–163.
Levelt, W. J. M. (1999). Producing spoken language: a blueprint of the speaker. Oxford:
Oxford University Press.
MacDonald, M. C., Pearlmutter, N. J., & Seidenberg, M. S. (1994). The lexical nature of
syntactic ambiguity resolution. Psychological Review, 101, 676–703.
McClelland, J., St. John, M., & Taraban, R. (1989). Sentence comprehension: A parallel
distributed processing approach. Language and Cognitive Processes, 4, 287–336.
McKinnon, R., & Osterhout, L. (1996). Constraints on movement phenomena in sentence
processing: Evidence from event-related brain potentials. Language and Cognitive
Processes, 11, 495–523.
McNeil, M., & Doyle, P. J. (2000). Reconsidering the hegemony of linguistic explanations
in aphasia: The challenge for the beginning of the millennium. Brain and Language,
71, 154–156.
Menon, R. S., & Goodyear, B. G. (2001). Spatial and temporal resolution of fMRI. In P.
Jezzard, P. Matthews, & S. Smith (Eds.), Functional MRI: An Introduction to Methods.
Oxford: Oxford University Press.
Milberg, W., & Blumstein, S. (2000). Back to the future: Reclaiming aphasia from cognitive
neurolinguistics. Brain and Language, 71, 160–163.
Mulert, C., Gallinat, J., Pascual-Marqui, R., Dorn, H., Frick, K., Schlattmann, P., et al. (2001).
Reduced event-related current density in the anterior cingulated cortex in schizophre-
nia. NeuroImage, 13, 589–600.
AU: Given as
in prepara-
tion in text
AU: Not cited
in text. Page 303 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
Nakagome, K., Takazawa, S., Kanno, O., Hagiwara, H., Nakajima, H., Itoh, K., et al. (2001).
A topographical study of ERP correlates of semantic and syntactic violations in the
Japanese language using the multichannel EEG system. Psychophysiology, 38,
Neville, H., Mills, D., & Lawson, D. (1992). Fractionating language: Difference neural
subsystems with different sensitive periods. Cerebral Cortex, 2, 244–258.
Neville, H. J., Nicol, J. L., Barss, A., Forster, K. I., & Garret, M.(1991). Syntactically based
sentence processing classes: Evidence from event-related brain potentials. Journal of
Cognitive Neuroscience, 3, 151–165.
Newman, A. J., Pancheva, R., Ozawa, K., Neville, H. J., & Ullman, M. T. (2001). An event-
related fMRI study of syntactic and semantic violations. Journal of Psycholinguistic
Research, 30, 339–364.
Ni, W., Constable, R. T., Mencl, W. E., Pugh, K. R., Fullbright, R. K., Schaywitz, B. A., &
Gore, J. (2000). An event-related neuroimaging study distinguishing form and content
in sentence processing. Journal of Cognitive Neuroscience, 12, 120–133.
Nunez, P. L. (1995). Neocortical dynamics and human EEG rhythms. Oxford: Oxford Uni-
versity Press.
Odlin, T. (1989). Language Transfer. Cambridge, UK: Cambridge University Press.
Osterhout, L. (1997). On the brain response to syntactic anomalies: Manipulations of word
position and word class reveal individual differences. Brain and Language, 59,
Osterhout, L. (2000). On space, time, and language: For the next century, timing is (almost)
everything. Brain and Language, 71, 175–177.
Osterhout, L., Allen, M., & McLaughlin, J. (2002). Words in the brain: Lexical determinants
of word-induced brain activity. Journal of Neurolinguistics, 15, 171–187.
Osterhout, L., Bersick, M., & McKinnon, R. (1997). Brain potentials elicited by words: Word
length and frequency predict the latency of an early negativity. Biological Psychology,
46, 143–168.
Osterhout, L., Bersick, M., & McLaughlin, J. (1997). Brain potentials reflect violations of
gender stereotypes. Memory and Cognition, 25, 273–285.
Osterhout, L., & Holcomb, P. J. (1992). Event-related brain potentials elicited by syntactic
anomaly. Journal of Memory and Language, 31, 785–806.
Osterhout, L., & Holcomb, P. J. (1993). Event-related potentials and syntactic anomaly:
Evidence of anomaly detection during the perception of continuous speech. Language
and Cognitive Processes, 8, 413–438.
Osterhout, L., & Holcomb, P. J. (1995). Event-related brain potentials and language compre-
hension. In M. D. Rugg & M. G. H. Coles (Eds.), Electrophysiology of mind: Event-
related brain potentials and cognition. Oxford: Oxford University Press.
Osterhout, L., Holcomb, P. J., & Swinney, D. A. (1994). Brain potentials elicited by garden-
path sentences: Evidence of the application of verb information during parsing.
Journal of Experiment Psychology: Learning, Memory, and Cognition, 20, 786-803.
Osterhout, L., McKinnon, R., Bersick, M., & Corey, V. (1996). On the language-specificity
of the brain response to syntactic anomalies: Is the syntactic positive shift a member
of the P300 family? Journal of Cognitive Neuroscience, 8, 507–526.
Osterhout, L., McLaughlin, J., Allen, M., & Inoue, K. (2002). Brain potentials elicited by
prose-embedded linguistic anomalies. Memory and Cognition, 30, 1304–1312.
Osterhout, L., McLaughlin, J., & Bersick, M. (1997). Event-related brain potentials and human
language. Trends in Cognitive Sciences, 1, 203–209.
Osterhout, L., & Mobley, L. A. (1995). Event-related brain potentials elicited by failure to
agree. Journal of Memory and Language, 34, 739–773.
AU: Not cited
in text.
AU: Not cited
in text. Page 304 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
Osterhout, L., & Nicol, J. (1999). On the distinctiveness, independence, and time course of
the brain responses to syntactic and semantic anomalies. Language and Cognitive
Processes, 14, 283–317.
Pascual-Marqui, R. D. (1999). Review of methods for solving the EEG inverse problem.
International Journal of Bioelectromagnetism, 1, 75–86.
Pascual-Marqui, R. D., Esslen, M., Kochi, K., & Lehmann, D. (2002). Functional imaging
with low resolution brain electromagnetic tomography (LORETA); review, new com-
parisons, and new validation. Japanese Journal of Clinical Neurophysiology, 30,
Pascual-Marqui, R. D., Michel, C. M., & Lehmann, D. (1994). Low resolution electromagnetic
tomography: A new method for localizing electrical activity in the brain. International
Journal of Psychophysiology, 18, 49–65.
Perani, D., Dehaene, S., Grassi, F., Cohen, L., Cappa, S. F., Dupoux, E., et al. (1996). Brain
processing of native and foreign languages. Neuroreport, 7, 2439–2444.
Perani, D., Paulesu, E., Galles, N. S., Dupoux, E., Dehaene, S., Bettinardi, V., et al. (1998).
The bilingual brain. Proficiency and age of acquisition of the second language. Brain,
121, 1841–1852.
Price, C. (1998). The functional anatomy of word comprehension and production. Trends i n
Cognitive Sciences, 2, 281–288.
Ringbom, H. (1987). The role of the first language in foreign language learning. Clevedon,
Somerset, UK: Multilingual Matters.
Rugg, M. D., & Coles, M. G. H. (1995). Electrophysiology of mind—Event-related brain
potentials and Cognition. Oxford: Oxford University Press.
Savoy, R. L., Bandettini, P. A., O’Craven, K. M., Kwong, K. K., Davis, T. L., Baker, J. R.,
et al. (1995). Pushing the temporal resolution of fMRI: Studies of very brief visual
stimuli, onset variability, and asynchrony, and stimulus-correlated changes in noise.
Proceedings of the Society of Magnetic Resonance, 2, 450.
Takazawa, S., Takahashi, N., Nakagome, K., Kanno, O., Hagiwara, H., Nakajima, H., et al.
(2002). Early components of event-related potentials related to semantic and syntactic
processes in the Japanese language. Brain Topography, 14, 169–177.
Tootell, R. B. H., Hadjikhani, N. K., Mendola, J. D., Marret, S., & Dale, A. (1998). From
retinotopy to recognition: fMRI in human visual cortex. Trends in Cognitive Sciences,
2, 174–183 .
Trueswell, J. C., & Tanenhaus, M. K. (1994). Toward a lexicalist framework of constraint-
based syntactic ambiguity resolution. In. C. Clifton, L. Franzier, & K. Rayner (Eds.),
Perspectives on Sentence Processing. Hillsdale, NJ: Erlbaum.
Wartenburger, I., Heekeren, H. R., Abutalebi, J., Cappa, S. F., Villringer, A., & Perani, D.
(2003). Early setting of grammatical processing in the bilingual brain. Neuron, 37,
Weber-Fox, C. M., & Neville, H. J. (1996). Maturational constraints on functional special-
izations for language processing: ERP and behavioral evidence in bilingual speakers.
Journal of Cognitive Neuroscience, 8, 231–256.
1. Some event-related designs are probably better than others if the goal is to isolate
the BOLD response to a specific word within a sentence. For example, a design in
which sentences are identical across conditions except for a single critical word, and
in which the hemodynamic response to critical words in successive sentences are
sufficiently separated in time, might be one effective means for isolating word-specific
BOLD responses. Also, recent efforts have been made to reduce the effective temporal
AU: Pls. pro-
vide location.
AU: Not cited
in text.
AU: Not cited
in text. Page 305 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
resolution (or “localization”) of fMRI signals (Menon & Goodyear, 2001). However,
whether these methods can be successfully applied to psycholinguistic research
remains to be seen.
2. These studies did differ in how the fMRI data were treated statistically, which by
itself could result in different conclusions across studies, even given similar BOLD
3. Although the N400 and P600 effects to semantic and syntactic processing, respec-
tively, are in fact highly reproduceable, and the antecedent conditions that elicit them
are fairly well understood, this is perhaps less so for the LAN and other anterior
negavitities elicited by syntactic anomalies. These anterior negativities are often not
observed under conditions when they should be, according to some theories (e.g.,
Friederici, 2002). We comment further on this issue in the “Caveats and Cautions”
4. We computed the LORETA solutions using a grand-average waveform comprised of
approximately 600 trials per condition (20 subjects times 30 trials per condition).
The waveform was projected onto the Montreal Neurological Institute averaged brain.
LORETA solutions were obtained for a single sample point representing the approx-
imate peak of each ERP effect of interest.
5. Despite these interesting consistencies, the LORETA solutions should be viewed as
preliminary until they are replicated. To the best of our knowledge, only two published
studies have attempted to localize the N400 component (or the magnetic equivalent)
using similar procedures (and none has attempted to localize the P600 effect). These
studies produced results that differ from each other and from our results. Haan, Streb,
Bien, and Roesler (2000) asked subjects to read pairs of words, some of which were
semantically related, and estimated the current distribution for the N400 on an indi-
vidual-subject basis. They found no consistent pattern across subjects. Dale et al.
(2002) asked subjects to read sentences with a sentence-final word that was either
fully acceptable or pragmatically implausible given the context. Source modeling
placed N400m (the magnetic equivalent) onset (at ~250 ms) in the left posterior
temporal lobe, and N400m peak (at 370 ms) in the left anterior orbital and frontopolar
cortices. There are numerous procedural differences across studies that could account
for these differences in outcome. Most fundamentally, both Haan et al. and Halgren
et al. computed solutions on difference waves (in which the responses to the fully
acceptable words were subtracted from the responses to the pragmatically anomalous
words). However, the difference waveform is a mathematical fiction that might not
preserve the underlying source information. Furthermore, Haan et al. presented word
pairs rather than sentences, and Halgren et al. presented the critical words at the ends
of sentences. It is not clear that these conditions will generalize to the conditions we
used, in which the critical words were embedded in sentences. For example, by placing
the critical words in sentence-final position, one potentially confounds the response
to the anomaly with sentence wrap-up effects (cf. Osterhout, 1997, for more on this
important point). Finally, Haan et al. and Halgren et al. both computed LORETA
solutions on individual subject data, whereas we chose to compute them using grand-
average data. Single-subject solutions probably preserve more precise spatial infor-
mation; however, they also contain fewer trials and therefore more noise. LORETA
cannot discriminate between signal and noise and models both; therefore, the validity
of the procedure depends on reducing the noise as much as possible. This is most
effectively done by using grand averages over subjects as the basis of the LORETA
procedure. Evaluations of LORETA’s reliability must take such methodological vari-
ation into account. Page 306 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
6. A reviewer suggested that the garden-path model could also account for these results.
Perhaps the N400 effect is elicited only when the final/eventual analysis is implau-
sible. It is conceivable that the implausible interpretation of (2a) is initiated, but
rapidly detected as implausible. This could trigger rejection of the initial semantic
interpretation before it is integrated along with an attempt to syntactically re-analyze
the sentence. The syntactic re-analysis may be manifested in the P600 effect at the
verb and the failure to integrate the initial interpretation may explain the absence of
an N400 effect. The difficulty with this explanation is that a vast literature shows that
N400 amplitude is exquisitely sensitive to pragmatic implausibility, and in fact is
roughly linearly related to gradations of plausibility. Furthermore, the N400 effect
typically onsets within ~250 ms after presentation of the anomalous word (which
leaves little time for additional processing to occur). Therefore, it seems highly
unlikely that implausibility would fail to be manifested in the N400 to the verb in
sentences like (2a) if the verb was perceived to be pragmatically implausible.
7. It is worth mentioning that the temporal ordering of the N400 and P600 is seemingly
inconsistent with the general assumption that syntactic processing precedes semantic
processing. Caution is needed in using the temporal qualities of these and other ERP
effects to make inferences about the timing of linguistic processes, due to the fact
that the specific cognitive events underlying these all ERP-sensitive effects are not
known. Nonetheless, it is interesting to note that linguistic (Jackendoff, 2000) and
psycholinguistic (Bever & Townsend, 2001) theories have disputed the syntactocentric
and syntax-first assumptions so prevalent in the field.
8. Several recent studies have examined ERP responses to linguistic anomalies presented
in an L2 (e.g., Hahne & Friederici, 2001; Weber-Fox & Neville, 1996). These studies
have in some cases produced results that are seemingly inconsistent with those
reported here. For example, Weber-Fox and Neville examined the ERP response to
linguistic anomalies in adult Chinese-English bilinguals who had been exposed to
English at various ages, ranging from 1 to 16 years. All subjects had lived in the U.S.
for a minimum of 5 years. ERPs were obtained in response to English pragmatic
anomalies and several types of syntactic anomalies, including phrase structure anom-
alies. All groups displayed an N400 effect to pragmatic anomalies. However, the
phrase structure anomalies elicited a P600-like effect only in bilinguals who were
exposed to English before the age of 16; for those subjects who were first exposed
to English after the age of 16, the phrase structure anomalies elicited an N400-like
response. However, there are many differences between the Weber-Fox and Neville
study and the study reported here, including language similarity (English-French vs.
Chinese-English), study design (longitudinal vs. cross-sectional), quality and quantity
of linguistic experience, the complexity of the stimuli, etc. We take our results as
indicating that it is possible to see the rapid development of native-like brain responses
to both pragmatic and syntactic anomalies given sufficient similarity between the
L1–L2, and especially L1–L2 similarity in the syntactic rules being tested.
9. We want to stress the distinction between claiming that some ERP effect correlates
highly with some stimulus manipulation and claiming that some ERP effect manifests
some particular process. For example, we have claimed and continue to believe that
the P600 effect is highly correlated with syntactic processing difficulty. We do not
think that there is good evidence that the P600 effect reflects specifically structural
reanalysis, as some have proposed.
10. In its most recent incarnations, Friederici’s theory is actually a three-stage theory:
the first stage is indexed by an “ELAN” effect between 100 and 300 ms, the second
stage by the LAN effect, and the third stage by the P600 effect. These effects are Page 307 Tuesday, March 23, 2004 4:49 PM
© 2004 by Taylor & Francis
claimed to reflect detection of a word category anomaly, detection of a morphosyn-
tactic anomaly, and syntactic reanalysis, respectively. We have reduced the stages to
two for ease of exposition; our reasoning applies equally well to a two- or three-stage
11. Predictions about what type of brain response should be elicited by this anomaly
depend largely on one’s theory of Japanese grammar. Having two nouns assigned the
same case (that is, the same thematic role) might be viewed as a syntactic problem
or a thematic–semantic problem. The uncertainty about how a Japanese speaker would
respond to these anomalies was one of the primary motivations for the experiment.
12. A simple statistical test exists for determining whether a biphasic LAN-P600 response
seen in a grand average waveform is biphasic or monophasic for individuals. If
subjects show either the LAN or the P600 (but not both) in response to a syntactic
anomaly, then the effect sizes (i.e., the amplitude difference between the well-formed
and ill-formed conditions) will be negatively correlated (indicating that as one effect
gets larger, the other gets smaller). In the Inoue and Osterhout study, the correlation
between the LAN and P600 effect sizes was robustly negative. This test can be applied
to any relevant data set. Page 308 Tuesday, March 23, 2004 4:49 PM
... In contrast, number agreement in English is poorly signaled due to a less detailed system of morphological markers, and speakers instead rely on word-order for sentence interpretation. Number agreement studies conducted with late L2 learners have shown evidence of cross-linguistic influence from the L1 onto the L2, revealing non-native-like processing profiles (i.e., missing LAN and/or P600) in cases where the L2 agreement properties in question did not exist in the speakers' L1 (e.g.,Chen, Shu, Liu, Zhao, & Li, 2007;Ojima et al., 2005;Osterhout, McLaughlin, Kim, Greenwald, & Inoue, 2004;Tokowicz & MacWhinney, 2005). Several studies have emphasized continued neuroplasticity by showing that adult L2 learners converge on native speakers' processing patterns with continued learning and high proficiency levels (e.g.,Hopp, 2010;Osterhout et al., 2006Osterhout et al., , 2008Rossi et al., 2006). ...
Full-text available
First language (L1) attrition in adulthood offers new insight on neuroplasticity and the role of language experience in shaping neurocognitive responses to language. Attriters are multilinguals for whom advancing L2 proficiency comes at the cost of the L1, as they experience a shift in exposure and dominance (e.g., due to immigration). To date, the neurocognitive mechanisms underlying L1 attrition are largely unexplored. Using event-related potentials (ERPs), we examined L1-Italian grammatical processing in 24 attriters and 30 Italian native-controls. We assessed whether (a) attriters differed from non-attriting native speakers in their online detection and re-analysis/repair of number agreement violations, and whether (b) differences in processing were modulated by L1-proficiency. To test both local and non-local agreement violations, we manipulated agreement between three inflected constituents and examined ERP responses on two of these (subject, verb, modifier). Our findings revealed group differences in amplitude, scalp distribution, and duration of LAN/N400 + P600 effects. We discuss these differences as reflecting influence of attriters’ L2-English, as well as shallower online sentence repair processes than in non-attriting native speakers. ERP responses were also predicted by L1-Italian proficiency scores, with smaller N400/P600 amplitudes in lower proficiency individuals. Proficiency only modulated P600 amplitude between 650 and 900 ms, whereas the late P600 (beyond 900 ms) depended on group membership and amount of L1 exposure within attriters. Our study is the first to show qualitative and quantitative differences in ERP responses in attriters compared to non-attriting native speakers. Our results also emphasize that proficiency predicts language processing profiles, even in native-speakers, and that the P600 should not be considered a monolithic component.
Research on second language acquisition has greatly benefitted from the declarative procedural model (DPM). The DPM proposes that declarative and procedural memory support – respectively – the acquisition of the L2 lexicon and L2 grammar. However, over the years, the meaning of ‘grammar’ and ‘lexicon’ within the DPM has changed, perhaps as a result of the changes occurred in linguistic theory. In particular, a number of studies coupled the combinatorial rules of grammar supported by procedural memory with patterns of usage and probabilistic rules. Such specification apparently excluded ‘noncombinatorial grammar’, which is the grammar computed over frequency-independent features. This paper first discusses the evolution of ‘grammar’ and ‘lexicon’ within the DPM, then describes a number of noncombinatorial phenomena that have been overlooked in memory studies on bilingualism. The aim is to encourage future research on whether and how especially procedural memory can also support the acquisition of noncombinatorial grammar.
Full-text available
Neuroergonomics focuses on the brain signatures and associated mental states underlying behavior to design human-machine interfaces enhancing performance in the cognitive and physical domains. Brain imaging techniques such as functional near-infrared spectroscopy (fNIRS) and electroencephalography (EEG) have been considered key methods for achieving this goal. Recent research stresses the value of combining EEG and fNIRS in improving these interface systems' mental state decoding abilities, but little is known about whether these improvements generalize over different paradigms and methodologies, nor about the potentialities for using these systems in the real world. We review 33 studies comparing mental state decoding accuracy between bimodal EEG-fNIRS and unimodal EEG and fNIRS in several subdomains of neuroergonomics. In light of these studies, we also consider the challenges of exploiting wearable versions of these systems in real-world contexts. Overall the studies reviewed suggest that bimodal EEG-fNIRS outperforms unimodal EEG or fNIRS despite major differences in their conceptual and methodological aspects. Much work however remains to be done to reach practical applications of bimodal EEG-fNIRS in naturalistic conditions. We consider these points to identify aspects of bimodal EEG-fNIRS research in which progress is expected or desired.
Bilingual speakers often switch between languages in conversation without any advance notice. Psycholinguistic research has found that these language shifts (or code-switches) can be costly for comprehenders in certain situations. The present study explores the nature of these costs by comparing code-switches to other types of unexpected linguistic material. To do this, we used a novel EEG paradigm, the Storytime task, in which we record readings of natural texts, and then experimentally manipulate their properties by splicing in words. In this study, we manipulated the language of our target words (English, Spanish) and their fit with the preceding context (strong-fit, weak-fit). If code-switching incurs a unique cost beyond that incurred by an unexpected word, then we should see an additive pattern in our ERP indices. If an effect is driven by lexical expectation alone, then there should be a non-additive interaction such that all unexpected forms incur a similar cost. We found three effects: a general prediction effect (a non-additive N400), a post-lexical recognition of the switch in languages (an LPC for code-switched words), and a prolonged integration difficulty associated with weak-fitting words regardless of language (a sustained negativity). We interpret these findings as suggesting that the processing difficulties experienced by bilinguals can largely be understood within more general frameworks for understanding language comprehension. Our findings are consistent with the broader literature demonstrating that bilinguals do not have two wholly separate language systems but rather a single language system capable of using two coding systems.
Full-text available
The present article examines the proposal that typology is a major factor guiding transfer selectivity in L3/Ln acquisition. We tested first exposure in L3/Ln using two artificial languages (ALs) lexically based in English and Spanish, focusing on gender agreement between determiners and nouns, and between nouns and adjectives. 50 L1 Spanish-L2 English speakers took part in the experiment. After receiving implicit training in one of the ALs (Mini-Spanish, N = 26; Mini-English, N = 24), gender violations elicited a fronto-lateral negativity in Mini-English in the earliest time window (200–500 ms), although this was not followed by any other differences in subsequent periods. This effect was highly localized, surfacing only in electrodes of the right-anterior region. In contrast, gender violations in Mini-Spanish elicited a broadly distributed positivity in the 300–600 ms time window. While we do not find typical indices of grammatical processing such as the P600 component, we believe that the between-groups differential appearance of the positivity for gender violations in the 300–600 ms time window reflects differential allocation of attentional resources as a function of the ALs’ lexical similarity to English or Spanish. We take these differences in attention to be precursors of the processes involved in transfer source selection in L3/Ln.
Full-text available
In this mini-review, I use event-related potential (ERP) studies to test the minimalist program (MP) prediction that organisms with the faculty of language cognitively process languagelike systems in a qualitatively distinct manner. I first discuss “languagelike” as a technical term defined by recursion criteria. From this definition and using a generative perspective, I show that certain domains of math and music can be considered languagelike. These domains are then used as case studies to test whether or not different languagelike systems are cognitively processed in a similar manner. This is done by investigating the elicitation of common language-related ERPs (namely, the left-anterior negativity (LAN), N400, and P600) in these languagelike systems. I show that these systems do indeed elicit the same language-related ERPs, supporting the claim that different languagelike systems are processed similarly. I then discuss discrepancies between these systems, as exemplified by the P3, and I provide plausible accounts for interpreting those results. I ultimately conclude that present data on the LAN, N400, and P600 disprove language-specificity but that languagelike-specificity remains plausible, and as yet there is no reason to reject MP’s prediction that languagelike systems are processed in a qualitatively distinct way.
Full-text available
Classifiers are essential elements between numerals and nouns in Mandarin (e.g. “one-touCL-elephant”), but whether they serve a semantic or functional/morphosyntactic role in relation to the accompanying noun has been heatedly debated in linguistics. Previous ERP research consistently supported the semantic view with findings of N400; however, the apparent meaning clash in mismatched classifier-noun pairing in these studies might render morphosyntactic processing undetected. We created two violation conditions to explore classifier-noun agreement: incongruent GE-noun combinations (replacing a specific classifier with the meaning-devoid general classifier, GE) and outright grammatical mistakes (missing a required classifier). With congruent combinations as the baseline, GE-noun combinations elicited a negativity effect strikingly similar to that induced by the grammatical violation condition in phrases (Experiment 1) and sentences (Experiment 2), indicating the involvement of morphosyntactic processing in classifier-noun agreement. The finding suggests that there is a middle ground for the linguistic debate over the nature of classifier selection in relation to nouns.
Full-text available
Recordings of event-related brain potentials (ERPs) provide a rich source of information about the cognitive systems supporting real-time language use. However, the interpretation of ERPs can be complicated by individual differences that aren't reflected in traditional analyses or visualizations. This is problematic, as failure to recognize important and systematic individual differences has in some cases led to inappropriate interpretations of ERP effects, with neurocognitive models of language comprehension sometimes being built on these inappropriate interpretations. In this chapter we review work, largely from our lab, on individual differences in ERP studies of language comprehension and discuss the promise of work on individual differences, as well as the challenges. In some cases individual differences in ERPs manifest themselves quantitatively (i.e., systematic differences in effect amplitudes), but in other more complex cases, qualitatively (i.e., different types of effects in different individuals). We will describe work on individual differences in morphosyntactic and semantic processing in both native and nonnative language processing, as well as multi-modal communication and higher-order pragmatic inferencing. In this last vein we will describe some nascent work done in our lab using unsupervised machine learning algorithms to better understand underlying patterns of qualitative individual differences in the processing of scalar implicatures. We conclude by laying out some challenges and suggestions for future work.
Full-text available
The role native language transfer plays in L2 acquisition raises the question of whether L1 constitutes a permanent representational deficit to mastery of the L2 morphosyntax and prosody or if it can eventually be overcome. Earlier research has shown that beginning and low intermediate Anglophone L2 French learners are insensitive to French morphosyntactic and prosodic constraints in using in situ pronouns transferred from the L1. The prosodic transfer hypothesis (PTH) proposes that native prosodic structures may be adapted to facilitate acquisition of L2 prosodic structure. Our study presents new evidence from three Anglophone advanced learners of L2 French that indicates ceiling performance for pronoun production (99% accuracy in 300 tokens over nine interviews) and grammaticality judgment (98% accuracy). This native-like performance demonstrates target French morphosyntax and prosody, built—as predicted by the PTH—by licensing pronominal free clitics in a new pre-verbal L2 position distinct from post-verbal L1. Furthermore, the learners’ data confirms accurate prosody by way of appropriate prominence patterns in clitic + host sequences, correct use of clitics with prefixed verbs, use of stacked pronouns, as well as correct prosodic alternations involving liaison and elision. These results counter impaired representation approaches and suggest early missing inflection may be overcome.
In this chapter we explore a controversial hypothesis about the nature of the syntactic processing system. The hypothesis, roughly stated, is that some of the procedures that create grammatical patterns in sentences are in an important sense indifferent to the content of the symbols they manipulate, in somewhat the same way that the procedures for long multiplication are indifferent to the numbers involved in the computation. The purpose of this exploration is to uncover the linguistic or cognitive bases of a phenomenon noted by many, including Edward Sapir, who described it in this way: All languages evince a curious instinct for the development of one or more particular grammatical processes at the expense of others, tending always to lose sight of any explicit functional value that the process may have had in the first instance, delighting, it would seem, in the sheer play of its means of expression .... This feeling for form as such, freely expanding along predetermined lines and greatly inhibited in certain directions by the lack of controlling types of patterning, should be more clearly understood than it seems to be... these submerged and powerfully controlling impulses to definite form operate as such, regardless of the need for expressing particular concepts or of giving consistent external shape to particular groups of concepts (1921, pp. 60–61).
This chapter addresses the limitations imposed by the functional magnetic resonance imaging (fMRI) technique and suggests some strategies to improve spatial and temporal resolution. It begins with the discussion of limits to spatial resolution in fMRI and shows that there are a number of factors to consider in achieving high spatial resolution with fMRI. The spatial resolution of the haemodynamics is also discussed. Since the haemodynamics are the source of the fMRI signal, the chapter also discusses how different sources contribute to the fMRI signal and reviews current opinion on how well localized the haemodynamic response actually is. It also reviews fMRI techniques that can be used to eliminate unwanted contributions to the measured signal and thereby increase the contrast-to-noise ratio of the fMRI signal directly related to submillimetre neural activity. Finally, the chapter shows an example of the high spatial resolution capabilities of fMRI in the human visual cortex.
It has long been recognized that aphasic patients display sentence comprehension deficits. Further analyses have indicated that these impairments are particularly apparent when correct comprehension rests on grammatical properties of language. Here, grammatical properties refer to syntactic and morphological. Thus, patients have particular difficulty understanding sentences when their meaning cannot be derived from lexical/semantic and real world constraints. The focus on syntactic impairments in these patients is understandable, given the difficulties that the patients have in comprehending sentences. There are separate representations for phonological, lexical, syntactic, and semantic information. Linguistic representations are considered as patterns of activation of either populations of units or as individual nodes. Every node has a resting state, rate of activation, a maximal level of activation, and a decay function which is reached over some temporal domain until it resumes its original resting state. These nodes are organized within a network like architecture such that the activation of a node may influence others through processes of spreading activation. The summation of activation is implicit in most characterizations of network and spreading activation models of lexical processing.