What is who violating? A reconsideration of Linguistic Violations in Light of Event-Related Brain Potentials

Marta Kutas

Departments of Cognitive Science and Neurosciences
University of California, San Diego

Robert Kluender

Department of Linguistics
University of California, San Diego

INTRODUCTION

     It is often a worthwhile exercise to step back from one's field of
endeavor and to think in broad terms about what its proper goals are, 
or perhaps what they should be.  It seems to us 
that the proper goals of ERP research into language comprehension 
should at minimum include enquiries into the following questions.

(1) How many independent factors contribute to our understanding of
natural language?

(2) What function relates all these factors to comprehension?
Undoubtedly this function is non-linear since upon careful analysis
almost every biological function is. 

(3) Are these factors specific to language, or do they cut across
cognitive domains?

(4) Which consequences of these factors can be consciously modified
(e.g. stopped or gated), and which are not under conscious control?

Although these remain at present long-range objectives of ERP language
research, there is nonetheless a slowly accumulating body of evidence 
that has begun to address issues in each of these areas.  At the same
time, over ten years of research have established beyond a doubt that 
ERPs are effective tools for the investigation of factors and processes 
involved in language comprehension.  A host of ERP measures, including 
amplitude, latency, and distribution, have been shown to vary, often 
systematically, with factors relevant to understanding.  The majority of
these studies have taken advantage of the fact that ERPs are 
particularly sensitive indices of violations of expectancy.

     Kutas and Hillyard (1980) first observed that following a 
violation of a semantic expectation at the end of a sentence, there 
was a negative component in the averaged waveform that was larger over 
posterior regions than over anterior regions, and over the right 
hemisphere than over the left hemisphere.  This monophasic negative 
wave onset around 200 msec. and was between 300 and 400 msec. in 
duration; it was called the N400, to reflect the fact that it was 
negative-going and peaked around 400 msec. post-stimulus (see 
Figure 1).  Since then, a number of studies have shown the reliability 
and robustness of this effect in a variety of paradigms, some of which 
will be reviewed in this chapter (for reviews see Kutas and Van 
Petten, 1988; Fischler, 1990; Halgren, 1990).

     Violations are thus known to be good vehicles for eliciting ERP
effects.  These can in turn be modulated by other factors in a way 
that makes it possible to ask questions about various processing 
models of language.  However, the fact that this is the case does
not imply that language ERP components are specific or unique 
reflections of violation.  For example, the N400 was originally 
viewed as a semantic violation detector.  That perception has 
persisted to the present day, even though several studies have since 
demonstrated that while the N400 is largest in amplitude following 
semantic violations, these are neither necessary nor sufficient to 
yield an N400 component.  In this chapter we highlight those studies 
which have demonstrated the sensitivity of the N400 to factors other 
than semantic violation.

     Additionally, we address the question of just what constitutes 
a violation in the first place.  Although the utility of violations 
in helping to decide between models of language processing has been 
amply demonstrated, this in no way implies or guarantees that 
researchers know exactly what isexpecte and therefore what is 
"violate.  Nor can we overlook the fact that even the description of 
the domain of a violation is theory-dependent.  We would therefore 
like to raise the possibility that violations are epiphenomenal,
 i.e. reducible to more fundamental properties of language, or even 
of cognition in general.  

     We set about doing this in the following way.  First we 
review the more general biological and cognitive factors known to 
influence N400 amplitude, latency, and distribution.  Then we take a 
look at some recent studies which have investigated the effects of 
violations of various sorts on the ERP record of sentence processing.  
Finally, we review studies which have indicated that the violations 
in question are perhaps best viewed as emerging from the interaction of 
a number of different variables, some of which may not be 
language-specific.


THE INFLUENCE OF BIOLOGICAL AND COGNITIVE FACTORS ON THE N400


     There is ample evidence indicating that the N400 and the cognitive
function that underlies its elicitation are modality independent.  N400 
effects have been observed in the visual modality in a variety of
languages, including written English (Kutas, Van Petten, and Besson, 
1988), Dutch (Gunter, Jackson, and Mulder, in press), French (Besson 
and Macar, 1987; Ardal, Donald, Meuter, Muldrew, and Luce, 1990),
Spanish (Kutas, Bates, Kluender, Van Petten, Clark, and Blesch, in 
preparation) and Japanese (Koyama, Nageishi, Shimokochi, Hokama, 
Miyazato, Miyatani, and Ogura, 1991), for visually presented signs 
of American Sign Language in congenitally deaf adults (Neville, 1985;
Kutas, Neville, and Holcomb, 1987), for line drawings terminating 
sentences presented visually (Kutas and Van Petten, 1990; Nigam, 
Hoffman, and Simons, 1990), and in the auditory modality for
both natural and metered speech (McCallum, Farmer, and Pocock, 1984; 
Herning, Jones, and Hunt, 1987; O'Halloran, Isenhart, Sandman, and
Larkey, 1988; Connolly, Stewart, and Phillips, 1990; Holcomb and 
Neville, 1990; Bentin, Kutas, and Hillyard, in press).  In all 
these situations, the relative difference between the ERP to a word
or picture that fits with its context and to one that does not is a 
monophasic negativity. 

     Modality independence, however, does not imply an insensitivity 
to the eliciting sensory modality.  Indeed, the onset latency, peak 
latency, duration, and laterality of both the raw N400 elicited by 
one condition as well as the N400 effect itself (i.e. the difference 
between two conditions) seem to vary systematically as a function of 
modality.  For example, the N400 effect in comprehension of speech 
generally onsets earlier and lasts longer than that observed during 
reading.  Insofar as asymmetries have been obtained, N400s in the 
visual and auditory modalities seem to exhibit reversed lateralities; 
that is, the effect is larger over the right hemisphere than over the 
left hemisphere during reading, but reversed during speech processing 
(Holcomb and Neville, 1990).

     Age and family history of handedness are other biologically 
determined factors that influence the N400.  For the same stimulus
sets, young adults generate larger and more peaked N400 waves than do 
older adults (Harbin, Marsh, and Harvey, 1984; Hamberger and Friedman,
1989; Gunter, Jackson, and Mulder, in press); there appears to be a 
linear reduction in N400 amplitude with age (Kutas, Mitchiner, and 
Iragui, in preparation).  Although there have been no studies of the 
N400 specifically and systematically devoted to determining the relation
between its laterality and handedness, some patterns have emerged.  
During reading, the N400 seems to be larger over the right hemisphere 
as long as subjects do not have a left-handed relative in the immediate 
family.  The N400s of subjects with left-handed family members are 
markedly less asymmetric (Kutas, Van Petten, and Besson, 1988; see 
Figure 2).  As mentioned above, this right predominance of the visual 
N400 appears to be reversed in the auditory modality.  Whether the 
laterality of the auditory N400 is also sensitive to family history of 
left-handedness is unknown.

     N400 amplitude is also influenced by a number of cognitive 
manipulations.  Within an experimental session, N400 amplitudes are 
larger for the first occurrence of a word or picture relative to the 
second occurrence (for a review see Van Petten, Kutas, Kluender, 
Mitchiner, and McIsaac, 1991).  This reduction in N400 amplitude 
with repetition interacts with frequency and congruity.  N400 
repetition effects are more pronounced for low frequency than high 
frequency words (Rugg, 1990; Van Petten, Kutas, Kluender, Mitchiner, 
and McIsaac, 1991) and for incongruous than congruous sentence-terminal
words (Besson, Kutas, and Van Petten, in press).  Both the semantic 
priming and repetition priming effects on the N400 are further subject 
to attentional manipulations.  Priming effects are either reduced 
or absent for words occurring outside the focus of attention (Holcomb,
1986; Nobre and McCarthy, 1987; Otten, Rugg, and  Doyle, submitted).
On the average, N400 amplitudes are also smaller for the less fluent 
language of a bilingual (Ardal, Donald, Meuter, Muldrew, and Luce, 
1990; Kutas, Bates, Kluender, Van Petten, Clark, and Blesch, in
preparation).

     Although most of the studies utilizing the N400 effect have 
focused on modulations of its amplitude, some data concerning its 
timing have also accumulated.  Specifically, the onset and peak 
latencies -- and on occasion the duration -- of the N400 have 
similarly proven to be susceptible to biological and cognitive 
influences.  Both onset and peak latencies are delayed with increasing 
age (Gunter, Jackson, and Mulder, in press; Kutas, Mitchiner, and 
Iragui, in preparation).  The rate of presentation of words in 
sentences also affects N400 latency.  Semantically anomalous
words completing sentences presented one word at a time (also known as 
rapid serial visual presentation orRSV) elicit N400s with a peak 
latency around 400 msec. when the presentation rate is 250 msec. or 
slower.  At a presentation rate of one word every 100 msec., both the 
onset and peak of the N400 effect are delayed by between 80 to 100 
msec. (Kutas, 1987).  We have also found a relationship between the 
timing of the N400 effect and a bilingual's fluency in his/her two 
languages; the N400 in the less fluent language peaks later and lasts 
longer than that in the more fluent language (Kutas, Bates, Kluender,
Van Petten, Clark, and Blesch, in preparation; see also Ardal, Donald,
Meuter, Muldrew, and Luce, 1990).  The size of the set of semantically 
related words from which a word is chosen appears to influence N400 
latency as well.  In semantic verification tasks, the N400s elicited 
by words that do not match a specified definition both start and peak 
earlier if the definition mentions a polar opposite rather than a 
category with a number of other members.  A possibly related effect 
is that involving the ordinal position of the eliciting word in a 
sentence.  The N400 elicited by a semantically anomalous word in the 
middle of sentences, where context does not fully constrain the 
occurrence of an ensuing word, peaks approximately 30 msec. later
than that elicited by an anomalous word at sentence end, where 
contextual constraint is greater (Kutas and Hillyard, 1983).


ERP STUDIES OF THE EFFECTS OF VIOLATION ON SENTENCE PROCESSING


    Let us now examine some ERP consequences of various violations 
within the domain of language.  Kutas and Hillyard (1983) investigated 
the ERP consequences of morphological violations embedded within 
written text.  Specifically, subjects were asked to read a number of 
different passages one word at a time in order to answer a series of 
multiple choice questions.  In addition to various semantic violations,
the text included sentences with morphological violations of the 
following kinds.

(a) ...as a turtle grows its shell grow too.... 
(b) ...all turtles have four leg and a tail....
(c) Turtles will spit out things they does not like to eat.
(d) ...they tend to been as cool as....

These morphological violations did not yield an N400 with the same
amplitude and distribution as that elicited by the semantic violations,
but there was nonetheless an enhanced negativity in the 300-500 
msec. range relative to control sentences, as well as a hint of an 
enhanced late positivity that carried over into the next epoch.  A
replication (Kutas, Bates, Kluender, Van Petten, Clark, and Blesch, in 
preparation) with another group of monolingual subjects showed that 
both the enhanced negativity and positivity were real effects, and that 
the different morphological violations were associated with different 
ERP changes.  Bilingual Spanish speakers reading similar materials in 
Spanish showed no negativity but an even more pronounced positivity 
Figure 3).

     Recently, Osterhout (1990) examined the ERPs associated with a 
number of different syntactic violations.  In one of his paradigms, 
ERPs were recorded while subjects read sentences containing 
"subcategorization violation; after each sentence, subjects were 
asked to make a decision as to its well-formedness (a so-called 
"grammaticality judgemen).

     On the assumption that words in the lexicon are assigned to various 
categories (e.g. nouns, verbs, adjectives, adverbs, etc.), verbs as a 
class can be further subdivided into subcategories on the basis of the 
complements that they take.  For example, some verbs take direct object 
complements while others do not.  Thus the verb persuade requires 
a direct object but cannot take a prepositional complement (e.g. John
persuaded me. vs. *John persuaded to me.), while the verb hope never 
takes a direct object but can take a prepositional complement (e.g. 
*They hoped a solution. vs. They hoped for a solution.).  More generally
it can be said that persuade belongs to that subcategory of verbs which 
takes a direct object but not a prepositional complement, while hope 
belongs to the subcategory of verbs that does not take a direct object 
but does take a prepositional complement.  Another way of saying this is 
that persuadesubcategorize for a direct object while hope 
"subcategorize for a prepositional complement.  Hence a verb 
subcategorization violation refers to the state of affairs that pertains 
when an expectancy set up by the verb for a certain constituent (such as 
a noun phrase direct object or a prepositional phrase) is not fulfilled, 
or in other words, is violated. 

     None of the sentences that Osterhout used in his experiment contained 
direct objects, but half the sentences contained verbs that require a 
direct object and thus constituted subcategorization violations, while 
the other half contained verbs that have no such requirement and were thus 
well-formed.  These sentences were of the following types.

(a)  The broker hoped to sell the stock.
(b)  The broker persuaded to the sell the stock.

Although both hope and persuade also take infinitival complements (i.e. 
to sell the stock), persuade still requires an immediately following 
direct object (e.g. The broker persuaded me to sell the stock.) while 
hope does not (e.g. *The broker hoped me to sell the stock.).  Osterhout 
recorded ERPs to the preposition to, which can under no circumstances 
introduce a direct object noun phrase and is thus unambiguously wrong 
when a direct object is required, as in the case of persuade.  Thus 
the ERP to the to following persuaded could be expected to show some 
sign of the subcategorization violation as soon as subjects became aware 
that the required direct object was missing, relative to the ERP elicited 
by the to following hoped, which requires no direct object.  

     And, indeed, the data in Figure 4 reveal that the ERP to the 
subcategorization violation in the persuade condition is more
positive from about 400 msec. on than the corresponding ERP in the 
control hope condition; Osterhout labelled this reponse the P600. 
Notice that the ERP elicited by the final word of the persuade 
condition containing the subcategorization violations sports an 
N400 relative to the control hope condition which contains no syntactic 
violations.  It remains an open question whether this N400 reflects 
the inability of subjects to find a suitable syntactic structure to 
assign to such sentences, an inability to make sense of them semantically, 
or both.  Also worth noting is the similarity between the P600 observed 
by Osterhout and Holcomb and the enhanced late positivity observed by 
Kutas, Bates, Kluender, Van Petten, Clark, and Blesch (in preparation) 
following morphological violations (see Figure 3).  More recording 
sites and greater scrutiny will be required to ascertain whether or 
not these two positivities are in fact equivalent.

     Recently Neville, Nicol, Barss, Forster, and Garrett (1991) reported on
the ERPs associated with violations of certain syntactic constraints 
posited by extended standard theory (Chomsky, 1973, 1977, 1981,
1986a, 1986b).  In one condition, Neville et al included phrase 
structure violations, which for present purposes can best be viewed as 
violations of normal word order.  Subjects were asked to make
grammaticality judgements on sentences like:

(a) The scientist criticized a proof of the theorem.
(b) The scientist criticized Max's of proof the theorem.

The ERPs in response to the preposition of were compared across the two 
conditions.  The preposition violates the expectancy for a noun or 
adjective to follow Max's in (3b), but is acceptable following the noun 
proof in (3a).  The violation in (3b) was associated with a number of 
ERP effects relative to (3a), including a pronounced negativity between 
300 and 500 msec. over temporal regions of the left hemisphere, and a 
more broadly and bilaterally distributed late positivity (see Figure 5). 

     Neville et al also investigated violations of a constraint referred 
to as thespecified subject conditio (Chomsky, 1973, 1977, 1981, 1986a,
1986b; Huang, 1982).  Without going into the details and history of this 
constraint within syntactic theory, we can simply say that it is considered 
responsible for the ungrammaticality of (4b) relative to a grammatical 
sentence like (4a).

(a) What did the scientist criticize a proof of ___ ?
(b) What did the scientist criticize Max's proof of ___ ? 

     The ERPs in Figure 6 show that the response to the word proof 
following Max's in (4b) is associated with an enhanced negativity 
in the region of the N1 component (around 100 msec.) as well as between 
300 and 500 msec. post-stimulus, primarily over anterior regions of the 
left hemisphere.  Although the earlier effect occurs perhaps too early 
to be an enhancement of the N1 per se, and may instead be a remnant of 
differential processing of the previous word, Neville et al take
both ERP effects to be specific indices of the violation of the 
specified subject condition.  They looked at other violations as well, 
but these two are sufficient to show the diversity of ERP componentry 
elicited by varioussyntacti violations.

    So far we have simply described the various ERP components apparently
associated with violations of expectancies or constraints within 
different language domains.  Within the domain of syntax, the variety 
of ERP responses that have been observed suggests that there is no 
unique index of syntactic violation.  Although in the next section we 
will show that the N400 is not a specific response to semantic anomaly, 
we first demonstrate how the fact that an N400 is reliably elicited
by semantic anomaly has been used to yield information about syntactic 
processing.

     An example of this can be seen in the work of Garnsey and colleagues 
(Garnsey, Tanenhaus, and Chapman, 1989).  Garnsey exploited the fact 
that N400s reliably occur in response to semantically incongruous words
to test alternative views of how people parse sentences, that is, how 
they figure out the syntactic relations between words in a sentence.
In English, determining the grammatical functions (e.g. subject, direct 
object, indirect object, etc.) of individual words in a sentence is
usually fairly simple, as this information is explicitly encoded in word
order.  However, for certain constructions word order cannot be used to 
assign grammatical functions.  For instance, in wh-questions, where the 
questioned element occurs at the beginning of a sentence (e.g. Which 
customer did the secretary call?), it is impossible to know what the 
grammatical function of the questioned element (Which customer) is
until later in the sentence.  Immediately upon reading or hearing 
Which customer, one cannot know whether it serves as the subject, 
object of the verb, or object of a preposition, etc.  To understand this
sentence the reader must keep this questioned element in working memory 
until its grammatical function becomes clearly specified.  

     Wh-questions represent one type offiller-ga construction.  The
questioned element or phrase (Which customer) is known as the 
"fille.  The empty position in the sentence where the filler would 
occur in anech question (The secretary called which customer?) is 
called thega, often indicated by a blank line (Which customer did 
the secretary call ___ ?).  In order to understand a sentence containing 
a filler-gap construction, a hypothetical parser must locate the gap 
and assign the filler to it.  This procedure is occasionally difficult 
or costly in terms of the allocation of mental resources because there
are no cues about gap location until some time after the filler is 
presented, and because sometimes there is more than one possible gap 
location.  For these reasons, there exists a temporary ambiguity as to 
the grammatical function of the filler while the sentence is being processed.

     Depending on one's theoretical bent, one can choose among several 
different procedures that a parser might follow to deal with the 
ambiguities of filler-gap constructions (Fodor, 1978).  At one extreme 
is the view that an attempt is made to assign the filler to the first 
possible gap location; this approach is called thefirst resor 
strategy.  At the other extreme is thelast resor strategy, which 
requires the parser to leave the filler unassigned until there is 
unambiguous information about where the true gap is, and only then 
assign the filler to the gap.  For the sentence Which customer did 
the secretary call [___] about ___ ?, the firstpotentia gap 
location occurs after call in square brackets, whereas the 
actual gap occurs after the word about.  These first and last 
resort strategy views of how a wh-question is parsed make very 
different predictions as to what mental processes occur immediately 
following the word call.

     Garnsey et al (1989) devised an electrophysiological test of 
these two alternative positions by constructing sentences with embedded 
wh-questions wherein the filler was either plausible or implausible: 

(a) The businessman knew which customer the secretary called ___ at home.
(b) The businessman knew which article the secretary called ___ at home. 

As a control, Garnsey also included simple declarative sentences which
either did or did not contain a semantically anomalous word, thereby 
providing a baseline against which N400 effects in the embedded wh-question
sentences could be compared. 

(a) The businessman knew whether the secretary called the customer at home.  
(b) The businessman knew whether the secretary called the article at home.  

The beauty of this design is that while the filler which article in (5b) 
is anomalous and will therefore elicit an enhanced N400 component according 
to both resort parsing strategies, the two make very different predictions 
as to exactly where in the sentence the N400 should appear.  If the first 
resort strategy is in effect, then which article violates a semantic 
expectancy after called and should elicit an N400 at that point.  On the 
other hand, if the parser is using a last resort strategy, then the 
filler which article should not elicit an N400 until the sentence ends 
and it is clear that there is no gap to which that filler can meaningfully 
be assigned. 

     Since the ERP elicited by called does indeed contain an N400 when
preceded by an implausible filler (which article) but not by a 
plausible one (which customer) (see Figure 7), the results support
a first resort strategy and argue against the last resort strategy 
outlined above.  They also argue against the existence of an autonomous 
syntactic processing module:  in such a model the parser would make the 
filler-gap assignment at the first possible gap, but would not evaluate 
its semantic plausibility until some later stage of processing when all 
potential gap sites had been eliminated from consideration. 

     It should be emphasized that the N400 is not a direct reflection of 
gap-filling, but rather a reflection of the incongruity that is either
a consequence of making an implausible filler-gap assignment or at least
of evaluating the possibility of that assignment.  There are many 
potential sources of information that could help the parser in making 
filler-gap assignment, and Garnsey's present research is aimed at 
determining whether the parser makes use of these different sources of 
information, and if so, when.


ARE LINGUISTIC VIOLATIONS REDUCIBLE TO OTHER FACTORS?


     Numerous experiments attest to the fact that while semantic anomalies 
may elicit the largest amplitude N400s, the response is not unique or 
specific to the presence of a semantic anomaly.  As early as 1984, Kutas 
and colleagues (Kutas and Hillyard, 1984; Kutas, Lindamood, and Hillyard,
1984) showed in a series of studies that semantically congruous words of 
low cloze probability elicit a larger N400 than semantically congruous 
words of high cloze probability.  This finding indicated that semantic 
anomaly is not a necessary condition for the elicitation of an N400; 
instead, N400 amplitude varies as an inverse function of word expectancy 
in a given context.  In addition, Kutas and her colleagues showed that N400 
amplitude is attenuated in response both to semantically anomalous 
words and to low cloze probability words when these are semantically 
related to high cloze probabilitybest completion (Figure 8).  This 
finding suggested that N400 amplitude is modulated by the degree to 
which a word is semantically primed by prior context.

     More recent studies have shown that semantic anomaly is not even a 
sufficient condition for the elicitation of an N400 response.  For 
example, Besson, Kutas, and Van Petten (in press) have shown that by 
the third repetition of a sentence containing a semantic anomaly, the 
N400 effect is eliminated (Figure 9).  Although a particular word may
render a sentence anomalous in isolation, the mere act of repetition 
increases the expectancy for that particular word to occur in the same 
context, thereby reducing N400 amplitude.  Clearly it is the 
variation in expectancy and not the violation per se that determines 
whether an N400 will be elicited by an anomalous word.

     N400s are in fact elicited by all words to varying degrees as 
determined by a number of factors; more general biological and cognitive 
factors were outlined in the second section of this chapter.  We now turn to 
factors more closely tied to sentence processing.  Among the most 
robust of these observations concerning N400 amplitude are:

(1) N400s are larger in response to open-class words than to closed-class 
    words (Kutas and Hillyard, 1983; Kutas, Van Petten, and Besson, 1988; 
    Van Petten and Kutas, 1991a, 1991b; Besson, Kutas, and Van Petten, 
    in press).

(2) N400s are larger in response to open-class words of low frequency than 
    to open-class words of high frequency when such words occur in word 
    lists or at the beginning of sentences.  This effect of frequency on the
    400 interacts with semantic constraints (Van Petten and Kutas, 1991a, 
    1991b), semantic congruity (Besson, Kutas, and Van Petten, in press), 
    and word repetition (Rugg, 1990; Van Petten, Kutas, Kluender, Mitchiner
    and McIsaac, 1991; Besson, Kutas, and Van Petten, in press).

(3) N400 amplitudes are not only reduced by the prior occurrence of a
    semantic associate or semantically related word, but increasingly 
    dampened by the accumulation of semantic constraints such as those 
    that build up during the course of a sentence (Van Petten, 1989; Van
    Petten and Kutas, 1991a, 1991b).  The latter effect is manifest in a 
    relative reduction in the amplitude of N400s elicited by open-class 
    words as a function of word position in isolated but meaningful 
    sentences.  In two studies (Van Petten, 1989; Van Petten and Kutas, 
    1991a, 1991b), it has also been shown that this effect of contextual 
    constraint on N400 amplitude interacts with the eliciting word's 
    frequency of occurrence in the language such that the accumulation of 
    prior context overrides the initial difference in N400 amplitude 
    between high and low frequency words.


     It was with some of these observations in hand that Van Petten
used the N400 to focus on interactions between lexical and sentence-level 
factors in an attempt to resolve the controversy that revolves around
the time course of sentential context effects.  In a series of experiments, 
Van Petten and Kutas (1987, 1991a, 1991b) tested the timing predictions of 
two different views of language processing.  One view assumes that each 
word has a pre-defined lexical representation that is first accessed and 
only subsequently subject to higher level (e.g. sentential or discourse) 
constraints (Garrett, 1978; Forster, 1981; Fodor 1983); the other view is 
based on the assumption that a reader's primary goal is comprehension, 
and that the brain substrates for language are organized so that this goal 
of comprehension pervades and impacts other levels of analysis at a number
of different time points, both early and late (Marslen-Wilson 1987; 
Marslen-Wilson and Tyler, 1980, 1987; Tyler and Marslen-Wilson 1977; Tyler 
and Wessels, 1983).  On the whole, the results of the Van Petten and Kutas
studies cannot be accommodated easily by word processing models that 
rely strictly on static representations of words in a mental lexicon 
slowly built over years of use, and have led to our belief that 
comprehension is a product of converging constraints that operate in 
parallel at multiple levels of linguistic analysis.

     The main conclusions that can be drawn from these experiments include:

(1) the process which yields frequency effects for words presented 
    in isolation is not mandatory or immune to sentence-level context;

(2) the influence of sentence-level context can be as powerful and act as 
    early as that of a single lexically associated word; and 

(3) sentence context can be used to pick out the appropriate core meaning of
    an ambiguous word without first passing through an early stage of 
    indiscriminate semantic activation.

     Let us examine the logic of one of these experiments.  The materials for
this experiment consisted of four types of sentences, each containing a 
critical pair of words.  The same associated pairs of words (such as 
salt-pepper) were embedded in both congruent and syntactically legal but
semantically anomalous sentences Congruent Associate andAnomalous 
Associate, respectively).  Likewise, pairs of words which were not 
particularly related to one another outside of a sentence context were
embedded in both congruent and anomalous sentences Congruent
Unassociate andAnomalous Unassociate).  With this design, the
second word of a pair could thus benefit from both sentential and 
lexical context as in the Congruent Associated condition, lexical context 
alone as in the Anomalous Associated condition, sentential context alone as
in the Congruent Associated condition, and neither as in the Anomalous 
Unassociated condition.

    The ERPs elicited by the critical word pairs are shown in Figure 
10.  Based on what we know about the sensitivity of the N400 to lexical 
associative effects and to contextual effects, all but the Anomalous 
Unassociated condition should show an N400 effect.  The critical 
question, however, concerns the timing of the N400 effect.  Does the 
lexical N400 effect start earlier than the sentential N400 effect as the
autonomous lexical view would predict, or is there some other pattern of
timing relations?  The answer is that there is no indication that
the lexical effect precedes the sentential effect, for the N400 effects
appear to start at the same time.  Thus we find no evidence for the 
view that there is a distinct stage in the processing of a word that is
influenced by associative links but blind to sentence-level context. 

    These ERP data indicate that lexical and sentential effects are 
remarkably similar.  Indeed, there seems to be no reason to assume that 
the mechanisms that underlie the two are completely orthogonal.  The 
sensitivity of the N400 to lexical factors such as word class membership
and frequency as well as to sentential influences underscores its 
independence from semantic violations.

     This raises the question of whether the ERP effects seen in response to
so-called syntactic violations in the preceding section are actually 
specific responses to violations.  If they are, then of course it is 
necessary to figure out exactly what constraints are being violated, and
to show that a particular violation will always elicit a specific ERP 
pattern.  But consider the possibility, based on what we have learned 
about the N400 and its relation to semantic violations, that these ERPs 
are not specific to syntactic violations per se but rather reflect the 
interactions of various lexical, semantic, and pragmatic factors with 
the surface structural properties of language.  On this view, syntactic 
violations merely represent one end of a continuum of such interactions.  
At the opposite end of the continuum one would find interactions of these 
properties that produce completely grammatical sentences; in between 
these two extremes would exist an entire range of sentences whose 
grammaticality would be determined by the confluence of these factors in
various combinations.

     An example of this perspective on the relation between ERPs and
grammaticality is provided by the work of Kluender (Kluender,
1991; Kluender and Kutas, in preparation a).  Specifically, Kluender
focussed on ERPs elicited by words in yes/no- and wh-questions, making 
direct comparisons between sentences of both types that differed in the 
closed-class item that introduced an embedded clause (e.g. that, if, or
who/what).  

(a) Can't you remember *that* he advised them against it on previous occasions?
(b) Can't you remember *if* he advised them against it on previous occasions?
(c) Can't you remember *who* he advised ___ against it on previous occasions?
(d) What did you remember *that* he advised them against ___ on previous 
    occasions? 
(e) What can't you remember *if* he advised them against ___ on previous 
    occasions?
(f) What can't you remember *who* he advised ___ against ___ on previous 
    occasions?

This work was based on the notion that the membership of a word in the 
open or closed class is highly correlated with its frequency of 
occurrence in the language (Gentner, 1981), and with the extent to 
which it refers to entities or relations in the real world, i.e. its 
referentiality (Kluender, in press).  Kluender predicted that closed-class 
items would elicit N400s of different amplitude to the extent that they 
varied along these dimensions.  And indeed, as seen in Figure 11, which 
compares the yes/no-question conditions (7a) through (7c), who, the least
frequent and most referential of these closed-class words, elicits the 
largest N400; if, which is intermediate on both dimensions, elicits an N400 
intermediate in amplitude; and that, the most frequent and most semantically 
neutral item, elicits the smallest N400. 

     In addition to these strictly lexical ERP effects, Kluender and Kutas
(in preparation b) observed an ERP effect that reflected the structural 
properties of sentences containing filler-gap constructions.  Specifically,
any condition containing a filler-gap relationship elicits a larger negativity 
over left anterior regions when compared to conditions in which this 
filler-gap relationship is absent.  This can be seen in comparisons made at 
the subject he in the embedded clause of all three yes/no-questions 
(conditions 7a, 7b, and 7c):  when he is preceded by the embedded filler 
who as in condition (7c), a left anterior negativity is seen (Figure 12).  
Note that all three sentences are perfectly grammatical, i.e. no 
syntactic violation is involved in this comparison. 

     Figure 13 shows similar effects seen at three different positions
within the embedded clauses of wh-questions, conditions (7d),
(7e), and (7f).  Although all three conditions contain main clause 
wh-fillers associated with prepositional object gaps in the embedded 
clause, condition (7f) contains an additional filler-gap relationship 
in the embedded clause which does not exist in the other two conditions. 
Left anterior negativity is seen at the embedded subject he in condition
(7f), the wh-question with the embedded filler who, relative to conditions 
(7d) and (7e), which contain embedded that- and if-clauses.  Note that 
although the eliciting condition (7f) is in this case ungrammatical, the 
response does not differ in latency or distribution from that seen at the
same position in the grammatical yes/no-question (7c) in Figure 12.  
Furthermore, for some speakers condition (7e) is also considered 
ungrammatical, but here no left anterior negativity emerges in the ERP.

     The second position where this effect is seen is the embedded 
preposition against; in condition (7f), which contains the 
embedded filler-gap relationship, this preposition is preceded by a 
direct object gap, while in conditions (7d) and (7e) the preposition is 
preceded by a lexical noun phrase, in this example the pronoun 
them.  Once again, condition (7f) shows increased left anterior 
negativity relative to the other two conditions, this time in response 
to gap location.  Although the eliciting condition is again the 
ungrammatical one, a similar effect is seen in response to
the embedded preposition in condition (7c), i.e. in grammatical 
yes/no-questions with embedded direct object gaps.

     The third position is the preposition on in the embedded clause.
In all three conditions (7d), (7e), and (7f), this preposition is 
preceded by a prepositional object gap associated with the main clause 
filler What.  When all three conditions are thus equivalent in the
presence of an immediately preceding gap, lexical effects reflecting
the occurrence of who, if, or that earlier in the sentence modulate 
the amplitude of the negativity indexing the filler-gap relationship
(the possibility that the effect is due to the difference in the main 
clause auxiliaries did vs.can't is relatively remote; see Kluender 1991 
for discussion). The amplitude of the left anterior response is larger 
in conditions (7e) and (7f), which contain who and if embeddings, 
respectively, than in condition (7d), which contains a that embedding. 
The important points to note are (1) that this negativity has persisted
throughout the extent of the filler-gap domain, that is, until the 
filler is assigned and the dependency released, and (2) that
this structurally influenced sentential effect also responds to lexical
influences.  Lexically induced modulations of the response to 
sentence-final words (occasions) were also observed.

     Traditionally, the differences in the grammaticality of these sentences 
have been explained in terms of violations of syntactic constraints.  However,
it may be possible to explain them in terms of the processing requirements of
a filler-gap construction and the demands of processing intervening lexical 
items.  For example, these processing difficulties may reflect strain on the 
limited capacity of a working memory which must store a filler along with the 
intermediate products of comprehension.

CONCLUSION


    We hope to have convincingly laid to rest the misconception that the
N400 is exclusively associated with semantic anomaly; we reiterate for
emphasis that semantic anomaly is neither a necessary nor a sufficient 
condition for the elicitation of an N400.  Our reading of the currently
available research is that sentential processes can and do influence 
lexical processing, and conversely, that lexical processes influence 
sentential processing.  In other words, we do not see ERP evidence for 
a strict modularist position in these domains.  In addition, we have 
suggested that the canonical construal of linguistic violation may be 
fundamentally flawed.  Just as semantic anomalies have proved to be
derivable from an interaction of lexical properties (such as frequency
of occurrence and lexical class membership) with contextual constraints 
at the sentence and discourse levels, so too it may prove possible to
derive syntactic violations in the traditional sense from the
interaction of the sequential structure of language with lexical and 
semantic properties such as frequency of occurrence and referentiality.
If this is the case, then the next logical question is whether and to 
what extent such interactions can be found in other cognitive domains 
as well.  Memory has frequently been invoked as a factor influencing 
language processing (Carpenter and Just, in press; King and Carpenter,
in press, in preparation).  Thus if specified subject condition violations
can be shown to fall out from the interaction of short-term memory 
constraints with the need to access mental representations in simultaneous 
lexical processing, one would want to know if similar effects can be 
found in other memory tasks that do not involve language, say for 
example in perceptual monitoring tasks.  So in the long run, these data 
can be used to assess the common notion that language processing is
somehow special, unique among cognitive processes, different in the nature 
of the mental operations involved, and subject to very different processing 
constraints than for example face recognition or comprehension of 
visual scenes.  We are hopeful that this is an empirical question to
which answers can be found.

ACKNOWLEDGEMENTS


Marta Kutas and some of the research herein were supported by an RSDA
from NIMH (MH00322) and grants from NICHD (HD22614) and NIA (AG08313).
Robert Kluender was supported by a McDonnell-Pew postdoctoral fellowship
in cognitive neuroscience.  We would like to thank Claudia Brugman, 
Heather McIsaac, Ken Paller, and Cyma Van Petten for comments on earlier
versions of this chapter.

REFERENCES


Ardal S, Donald MW, Meuter R, Muldrew S, and Luce M (1990): Brain responses 
	to semantic incongruity in bilinguals. Brain and Language 39: 187-205.

Bentin S, Kutas M and Hillyard SA (in press): Electrophysiological evidence 
	for task effects on semantic priming in auditory word processing. 
	Psychophysiology.

Bentin S, McCarthy G, and Wood CC (1985): Event-related potentials associated
	with semantic priming. Electroencephalography and Clinical 
	Neurophysiology 60: 343-355.

Besson M and Macar F (1987): An event-related potential analysis of incongruity
	in music and other non-linguistic contexts.  Psychophysiology 24: 14-25.

Besson M, Kutas M and Van Petten C (in press): An event-related potential (ERP)
	analysis of semantic congruity and repetition effects in sentences. 
	Journal of Cognitive Neuroscience. 

Boddy J (1981):  Evoked potentials and the dynamics of language processing.
	Biological Psychology 13: 125-140.

Boddy J (1986): Event-related potentials in chronometric analysis of primed 
	word recognition with different stimulus onset asynchronies.
	Psychophysiology 23: 232-245.

Boddy J and Weinberg H (1981): Brain potentials, perceptual mechanisms and 
	semantic categorization. Biological Psychology 12: 43-61.

Carpenter PA and Just MA (in press): A capacity theory of comprehension:
	individual differences in working memory. Psychological Review.

Chomsky N (1973): Conditions on transformations. In: A Festschrift for 
	Morris Halle, Anderson S and Kiparsky P, eds. New York: Holt, 
 	Rinehart and Winston.

Chomsky N (1977): On wh-movement. In: Formal Syntax, Culicover P, Akmajian A 
	and Wasow T, eds. New York: Academic Press.

Chomsky, N. (1981). Lectures on Government and Binding. Dordrecht: Foris 
	Publications.

Chomsky, N. (1982). Some Concepts and Consequences of the Theory of Government 
	and Binding. Cambridge, MA: MIT Press.

Chomsky, N. (1986a). Barriers. Cambridge, MA: MIT Press.

Chomsky, N. (1986b). Knowledge of Language: Its Nature, Origin, and Use. 
	New York: Praeger.

Connolly JF, Stewart SH, and Phillips NA (1990): The effects of processing 
	requirements on neurophysiological responses to spoken sentences.
	Brain and Language 39: 302-318.

Fischler I (1990): Comprehending language with event-related potentials. 
	In: Event-Related Brain Potentials: Basic Issues and Applications,
	Rohrbaugh JW, Parasuraman R, and Johnson RE Jr, eds.  New York: 
	Oxford University Press. pp.165-177.

Fischler I, Bloom PA, Childers DG, Roucos SE and Perry NW Jr (1983):  
	Brain potentials related to stages of sentence verification.
	Psychophysiology 20: 400-409.

Fischler I, Bloom PA, Childers DG, Arroyo AA and Perry NW Jr (1984):
	Brain potentials during sentence verification: Late negativity
	and long-term memory strength.  Neuropsychologia 22: 559-568.

Fischler I, Childers DG, Achariyapaopan T and Perry NW Jr (1985): Brain 
	otentials during sentence verification: Automatic aspects of
	comprehension. Biological Psychology 21: 83-106.

Fodor JA (1983): The Modularity of Mind.  Cambridge, MA: MIT Press.

Fodor JD (1978): Parsing strategies and constraints on transformations.
	Linguistic Inquiry 9: 427-473.

Forster KI (1981): Priming and the effects of sentence and lexical contexts 
	on naming time: Evidence for autonomous lexical processing.
	Quarterly Journal of Experimental Psychology 33A: 465-495.

Garnsey SM, Tanenhaus MK and Chapman RM (1989): Evoked potentials and the 
	study of sentence comprehension.  Journal of Psycholinguistic 
	Research 18: 51-60.

Garrett M (1978): Word and sentence perception. In: Handbook of Sensory 
	Physiology 8, Perception, Held R, Leibowitz HW and Teuber HL eds. 
	Berlin: Springer-Verlag. (pp 611-626) 

Gentner D (1981): Some interesting differences between verbs and nouns. 
	Cognition and Brain Theory 4: 161-178.

Gunter TC, Jackson JL and Mulder G (in press): An electrophysiological
	study of semantic processing in young and middle aged academics.
	Psychophysiology.

Halgren E (1990): Insights from evoked potentials into the neuropsychological 
	mechanisms of reading. In: Neurobiology of Higher Cognitive Function,
	Scheibel AB and Wechsler AF, eds.  New York: Guilford Press. 
	pp. 103-149.

Hamberger MJ and Friedman D (1989): Age-related changes in semantic activation: 	Evidence from event-related potentials. Proceedings of EPIC IX, 
	Noordwijk, The Netherlands III: 38-39.

Harbin TJ, Marsh GR and Harvey MT (1984): Differences in the late components of
	the event-related potential due to age and to semantic and non-semantic 	tasks.  Electroencephalography and Clinical Neurophysiology 59: 489-496.

Herning RI, Jones, RT and Hunt JS (1987): Speech event related potentials 
	reflect linguistic content and processing level.  Brain and Language 
	30: 116-129.

Holcomb PJ (1985): Unimodal and multimodal models of lexical memory: an ERP 
	analysis (abstract).  Psychophysiology 22: 576. 

Holcomb PJ (1988): Automatic and attentional processing: An event-related brain 	potential analysis of semantic processing.  Brain and Language 35: 66-8

Holcomb PJ (1986): ERP correlates of semantic facilitation. In: 
	Electroencephalography and Clinical Neurophysiology Supplement 38,
	Cerebral Psychophysiology: Studies in Event-Related Potentials,
	(pp 320-322), McCallum WC, Zappoli R and Denoth F, eds. Amsterdam: 
	Elsevier.

Holcomb PJ and Neville HJ (1990): Auditory and visual semantic priming
	in lexical decision -- a comparison using event-related brain 
	potentials.  Language and Cognitive Processes 5: 281-312.

Huang CTJ (1982): Logical Relations in Chinese and the Theory of Grammar. 		Doctoral dissertation, MIT.

King J and Just MA (in press): Individual differences in syntactic processing:
	the role of working memory. Journal of Memory and Language.

King J and Just MA (in preparation): Individual differences in working 
	memory and the use of contextual information during parsing.

Kluender R (1991): Cognitive Constraints on Variables in Syntax.  Doctoral 
	dissertation, University of California, San Diego.

Kluender R (in press): Deriving island constraints from principles of 
	predication. In: The Psycholinguistics of Island Constraints,
	Goodluck H and Rochemont M, eds. Dordrecht: Kluwer Academic Press.

Kluender R and Kutas (in preparation a): The interaction of lexical and
	syntactic effects in the processing of unbounded dependencies.

Kluender R and Kutas (in preparation b): Filling the gap: Evidence from 
	ERPs on the processing of unbounded dependencies.

Koyama S, Nageishi Y, Shimokochi M, Hokama M, Miyazato Y, Miyatani M, 
	and Ogura C (1991): The N400 component of event-related potentials in 
	schizophrenic patients: A preliminary study.  Electroencephalography 
	and clinical neurophysiology 78: 124-132.

Kutas M (1981): Semantic processing and ERPs: A case study. Paper presented at
	the Third Annual Carmel Conference on ERPs, Carmel, California.

Kutas M (1983): Cognitive models based on ERP data. Tutorial presented at the 
	Sixth Annual Workshop on Cognitive Psychophysiology, Bellagio, Italy.

Kutas M (1984): Language processing and the N400 wave. Paper presented 
	at the Third International Conference on Cognitive Neuroscience,
	Bristol, England.

Kutas M (1985a): ERP comparisons of the effects of single word and 
	sentence contexts on word processing (abstract).Psychophysiology
	22: 575-576.

Kutas M (1985b):  Event-related brain potentials in bilinguals during 
	reading.  Paper presented at the Thirteenth Annual Meeting of the 
	International Neuropsychological Society, San Diego, California.

Kutas M (1987): Event-related brain potentials (ERPs) elicited during rapid 
	serial visual presentation of congruous and incongruous sentences. 
	In: Electroencephalography and Clinical Neurophysiology Supplement 40,
	Current Trends in Event-Related Brain Potential Research, pp. 406-411.

Johnson R Jr, Parasuraman R and Rohrbaugh JW, eds. Amsterdam: Elsevier.

Kutas M, Bates E, Kluender R, Van Petten C, Clark V and Blesch F (in
	preparation): What's critical about the critical period?  Effects
	of early experience in the language processing of bilinguals.

Kutas M and Hillyard SA (1980): Reading senseless sentences: Brain potentials 
	reflect semantic incongruity.  Science 207: 203-205.

Kutas M and Hillyard SA (1980b):  Event-related brain potentials to 
	semantically inappropriate and surprisingly large words. Biological 
	Psychology 11: 99-116.

Kutas M and Hillyard SA (1982): The lateral distribution of event-related 
	potentials during sentence processing. Neuropsychologia 20: 579-590.

Kutas M and Hillyard SA (1983): Event-related brain potentials to grammatical
	 errors and semantic anomalies.  Memory and Cognition 11: 539-550.

Kutas M and Hillyard SA (1984): Brain potentials during reading reflect 
	word expectancy and semantic association. Nature 307: 161-163.

Kutas M and Hillyard SA (1988): An electrophysiological probe of 
	incidental semantic association. Journal of Cognitive Neuroscience
	1: 38-49. 

Kutas M, Lindamood T and Hillyard SA (1984): Word expectancy and
	event-related brain potentials during sentence processing. In:
	Preparatory States and Processes, Kornblum S and Requin J, eds.
	(pp 217-238). New Jersey: Erlbaum Press.  

Kutas M, Mitchiner M and Iragui V (in preparation): The processing of
	semantic anomalies as indexed by N400s throughout the decades.

Kutas M, Neville HJ, and Holcomb PJ (1987): A preliminary comparison of 
	the N400 response to semantic anomalies during reading, listening, 
	and signing. In: Electroencephalography and Clinical Neurophysiology
	Supplement 39, The London Symposia, .(pp 325-330), Ellingson RJ, 
	Murray NMF and Halliday AM, eds. Amsterdam: Elsevier.

Kutas M and Van Petten C (1988): Event-related brain potential studies
	of language. In: Advances in Psychophysiology 3, Ackles PK, Jennings 
	JR and Coles MGH, eds. Greenwich, CT: JAI Press.

Kutas M and Van Petten C (1990): Electrophysiological perspectives on 
	comprehending written language. In: Electroencephalography and 
	Clinical Neurophysiology Supplement 41, New Trends and Advanced 
	Techniques in Clinical Neurophysiology, Rossini PM and Mauguiere F,
	eds. Amsterdam: Elsevier.

Kutas M, Van Petten C and Besson M (1988): Event-related potential asymmetries 
	during the reading of sentences.  Electroencephalography and Clinical 
	Neurophysiology 69: 218-233.

Marslen-Wilson WD (1987): Functional parallelism in spoken word-recognition. 
	Cognition 25: 71-102.

Marslen-Wilson WD and Tyler LK (1980): The temporal structure of spoken
	language understanding. Cognition 8: 1-71.

Marslen-Wilson WD and Tyler LK (1987): Against modularity. In: Modularity in 
	Knowledge Representation and Natural-Language Understanding, Garfield 
	JL, ed. Cambridge, MA: MIT Press.

Marslen-Wilson WD and Tyler LK (1980b): Towards a psychological basis
	for a theory of anaphora. In: Papers from the Parasession on
	Pronouns and Anaphora, Kreiman J and Ojeda AE, eds. Chicago: Chicago
	Linguistic Society.

McCallum WC, Farmer SF and Pocock PK (1984): The effects of physical 
	and semantic incongruities on auditory event-related potentials.
	Electroencephalography and Clinical Neurophysiology 59: 477-488.

Nagy ME and Rugg MD (1989): Modulation of ERP[????] by word repetition: The 
	effects of inter-item lag. Psychophysiology 26: .431-436.

Neville H (1980): Event-related potentials in neuropsychological studies
	of language. Brain and Language 11: 300-318.

Neville HJ (1985): Biological constraints on semantic processing: a 
	comparison of spoken and signed language (abstract). Psychophysiology
	22: 576.

Neville HJ, Kutas M, Chesney G and Schmidt A (1986): Event-related brain 
	potentials during initial encoding and subsequent recognition 
	memory of congruous and incongruous words. Journal of Memory and 
	Language 25: 75-92.   

Neville HJ, Nicol JL, Barss A, Forster KI and Garrett MF (1991): Syntactically 
	based sentence processing classes: Evidence from event-related brain 
	potentials. Journal of Cognitive Neuroscience 3: 151-165.

Nigam A, Hoffman JE and Simons XP (1990): N400 and semantic anomaly with
	words and pictures. Psychophysiology 27: S53.

Nobre AC and McCarthy G (1987): Visual selective attention to meaningful text: 
	An analysis of event-related potentials. Society for Neuroscience 
	Abstracts 13: 852.

O'Halloran JP, Isenhart R, Sandman CA and Larkey LS (1988): Brain responses to
	semantic anomaly in natural, continuous speech.  International Journal 
	of Psychophysiology 6: 243-254.

Osterhout L (1990): Event Related Brain Potentials Elicited During Sentence 
	Comprehension.  Doctoral dissertation, Tufts University.

Otten LJ, Rugg MD and Doyle MC (submitted): Modulation of event-related 
	potentials by word repetition: The role of selective attention.

Picton TW and Stuss DT (1984): Event-related potentials in the study of
	speech and language: a critical review. In: Biological Perspectives 
	on Language, (pp 303-360) Caplan DN, Lecours AR and Smith AM, eds.
	Cambridge, MA: MIT Press.

Polich JM, McCarthy G, Wang WS and Donchin E (1983): When words collide: 
	Orthographic and phonological interference during word processing.
	Biological Psychology 16: 155-180.

Rugg MD (1983a): The relationship between evoked potentials and lateral 
	asymmetries of processing. In: Tutorials in ERP Research: Endogenous 
	Components, Gaillard AWK and Ritter W, eds. (pp 369-383). Amsterdam: 
	North-Holland Publishing Company.

Rugg MD (1983b): Further study of the electrophysiological correlates of
	lexical decision. Brain and Language 19: 142-152.

Rugg MD (1984a): Event-related potentials in phonological matching tasks.
	Brain and Language 23: 225-240.

Rugg MD (1984b): Event-related potentials and the phonological processing of 
	words and non-words. Neuropsychologia 22: 435-443.

Rugg MD (1985): The effects of handedness on event-related potentials in a 
	rhyme-matching task.  Neuropsychologia 23: 765-775.

Rugg MD (1990): Event-related brain potentials dissociate repetition
	effects of high- and low-frequency words.  Memory and Cognition
	18: 367-379.

Sanquist TF, Rohrbaugh JW, Syndulko K and Lindsley DB (1980): Electrocortical 
	signs of levels of processing: perceptual analysis and recognition 
	memory. Psychophysiology 17: 568-576.

Tyler LK and Marslen-Wilson WD (1977): The on-line effects of semantic
	context on syntactic processing. Journal of Verbal Learning and
	Verbal Behavior 16: 645-659.

Tyler LK and Wessels J (1983): Quantifying contextual contributions to
	word recognition processes. Perception and Psychophysics 34: 409-420. 

Tyler LK and Marslen-Wilson WD (19??): Processing utterances in discourse 
	contexts: On-line resolution of anaphors. Journal of Semantics 1: 
	297-314. [YEAR MISSING]

Van Petten C (1989): Context effects in word recognition: studies with
	event-related brain potentials. Doctoral dissertation, University of
	California, San Diego.

Van Petten C and Kutas M (1987): Ambiguous words in context: and 
	event-related potential analysis of the time course of meaning 
	course of meaning activation. Journal of Memory and Language
	26: 188-208.

Van Petten C and Kutas M (1988): Semantic versus syntactic context and the
	word frequency effect. Psychophysiology 25: 487.

Van Petten C and Kutas M (1991a): Interactions between sentence context 
	and word frequency in event-related brain potentials.  Memory and 
	Cognition 18: 380-393.

Van Petten C and Kutas M (1991b): Influences of semantic and syntactic 
	context on open and closed class words.  Memory and Cognition
	19: 95-112.

Van Petten C and Kutas M (1991c): Electrophysiological evidence for
	the flexibility of lexical processing. In: Word and Sentence,
	Simpson G, ed. Amsterdam: North Holland Press. 129-174

Van Petten C, Kutas M, Kluender R, Mitchiner M and McIsaac H (1991):
	Fractionating the word repetition effect with event-related
	potentials. Journal of Cognitive Neuroscience 3: 131-150.

Woodward SH, Owens J, and Thompson LW (1990): Word-to-word variation in 
	ERP component latencies: spoken words.  Brain and Language
	38: 488-503.



FIGURES

FIGURE 1.

Grand average ERPs elicited by semantically congruous and incongruous
sentence-terminal words from frontal, central, and parietal midline
locations.  Negative is up in this and all subsequent figures.  Data
taken from Kutas and Hillyard (1980).

FIGURE 2.

Grand average ERPs for the first seven words of sentences that subjects 
read for meaning, averaged separately for the nine right-handed subjects
without, and the five right-handed subjects with left-handed 
relatives.  Right and left hemisphere sites are overlapped.  
Note the absence of asymmetry in the 300-600 msec. range for 
right-handers with left-handed family members.  Reprinted from Kutas, 
Van Petten and Besson (1988) by permission of Elsevier Publishers. 

FIGURE 3.

Grand average ERPs (n=15) from a group of Spanish/English bilinguals reading 
simple Spanish text for comprehenison.  Compared are ERPs elicited by verb 
tense violations and control words matched on word position, length, and 
lexical class. Spanish was the primary language of these bilinguals, who were 
first exposed to English after age eight.  Note the enhanced late positivity 
elicited by verb tense violations.

FIGURE 4.

Grand average ERPs (n=15) to subcategorization violations and control words 
(to) and to the final words of sentences containing such violations and their 
control words (stock).  Note the late positivity (P600) elicited by the 
subcategorization violations and the negativity (N400) elicited by sentence
terminal words in sentences containing the subcategorization violations.  Data 
taken from Osterhout (1990).

FIGURE 5.

Grand average ERPs (n=40) elicited by phrase structure violations and two 
subsequent words (dashed line) versus those elicited by control words (solid 
line).  Recordings are from temporal and occipital regions of the left and 
right hemispheres.  Adapted from Neville et al (1991). 

FIGURE 6.

Grand average ERPS (n=40) elicited by specified subject condition 
violations (dashed line) and control words (solid line) as subjects made
grammaticality judgements.  Words were presented once every 500 msec. 
Data taken from Neville et al (1991). 

FIGURE 7.

Grand average ERPs (n=32) for plausible (solid line) and implausible 
(dotted line) filler-gap sentences.  At the top of the figure are the 
sentence contexts preceding the words below to which ERPs were recorded.
Vertical dashed lines indicate the 200 msec. range across which N400 
amplitude was quantified.  The shading indicates the plausible-implausible 
difference.  Modified from Garnsey, Tanenhaus, and Chapman (1989). 

FIGURE 8.

A:  Examples of sentences used in this study.  Each pair of
sentences exhibits high contextual constraints, but whereas the first 
sentence in each pair is terminated by itsbest completio (as 
determined by a sentence completion norming study), the second
sentence in each pair is completed by a low cloze probability word. 
In the first pair the low cloze probability ending is semantically
unrelated to the best completion, and in the second pair it is
semantically related.  B:  Grand average ERPs from Pz for best 
completions (solid line) and for semantically related (large dashed 
line) and semantically unrelated (small dashed line) low cloze 
probability words.  Reprinted from Kutas and Hillyard (1984).

FIGURE 9.

Grand average ERPs (n=17) elicited by semantically congruous and 
incongruous sentence endings after three different presentations 
separated by over fifteen minutes.  Note the reduction of N400 amplitude
with repetition.  Data taken from Besson, Kutas and Van Petten (in 
press).

FIGURE 10.

Grand average ERPs to the critical pairs of words (shown capitalized) in
each of the four sentence types used to contrast sentential and lexical
contexts.  Reprinted from Van Petten and Kutas (1991c).

FIGURE 11.

Grand average ERPs (n=30) elicited by that- (solid line) and 
if-complementizers (dashed line) and by the wh-fillers who and what 
(dotted line) in the initial position of embedded clauses in yes/no-questions. 
Note the typical distribution of the monotonic N400 (also in the 500-700 
msec. range):  greater over the right hemisphere than over the left, and over
posterior regions than over anterior regions.  Words were presented every 600 
msec.; subjects were given a probe word task at the end of each sentence. 
Data taken from Kluender (1991).

FIGURE 12.

Grand average ERPs (n=30) recorded at three anterior left hemisphere 
sites to function words (capitalized) immediately following embedded 
complementizers and wh-fillers in yes/no-questions.  Note the enhanced 
negativity between 300 and 500 msec. associated with the existence of a 
filler-gap relationship in the condition containing an embedded wh-question. 
Data taken from Kluender (1991).

FIGURE 13.

Grand average ERPs (n=30) from three anterior left hemisphere sites
elicited by three different positions in wh-questions:  the function word 
immediately following embedded complementizers and wh-fillers (HE), the 
preposition immediately following the embedded direct object position 
(AGAINST), and the initial function word of a sentence-ending adverbial 
immediately following the prepositional object gap in all three conditions 
(ON).  Note that at the latter position, the left anterior negativity is 
modulated in much the same way as the monotonic N400 seen in Figure 11.  
Data taken from Kluender (1991).