Words in context: ERPs and the lexical/postlexical distinction
Seana Coulson and Kara D. Federmeier
Department of Cognitive Science
University of California, San Diego
Correspondence to:
Seana Coulson
Cognitive Science 0515
9500 Gilman Drive
La Jolla, CA 92093-0515
email: coulson@cogsci.ucsd.edu
fax: (858) 534-1128
phone: (858) 822-4037
ABSTRACT
Classic models of language
comprehension posit a distinction between lexical processing, thought to
be early, fast-acting, and automatic, and post-lexical processing, thought
to be slower, more controlled, and to require the output of the lexical
processing system for its operation.
The lexical/post-lexical distinction is reviewed in light of data from
event-related brain potentials (ERPs) collected from normal volunteers as they
read or listened to words in context.
The lexical/post-lexical distinction is undermined by reports that
variables assumed to influence lexical processing, including frequency,
repetition, association, and orthographic legality, variables assumed to
influence post-lexical processing, including predictability, and even variables
assumed to influence discourse-level processing, such as congruity with global
context, all similarly modulate a negative-going ERP component (N400) evident
between 200 and 500 ms after word onset.
The similarity in the time course of lexical and post-lexical ERP
effects argues against the temporal precedence of lexical processing. Data demonstrating that lexical effects can
be attenuated by contextual factors further argue against the automaticity of
lexical processing. The interaction of
low-level and higher-level manipulations of context suggests that word
processing occurs at multiple levels of analysis in parallel, such that lexical
access and contextual integration are to some extent interdependent.
Adult
humans process language so frequently, so quickly, and so effortlessly that it
is easy to forget how complicated the task really is. To comprehend a spoken sentence, for example, the listener must
first process a complicated sensory signal.
Auditory speech contains information in multiple frequency bands,
distributed across multiple time scales.
Millisecond-level differences in timing must be appreciated to
differentiate the consonants that, in turn, differentiate words like
"pat" and "bat." At
the same time, the listener must be able to recognize tonal patterns unfolding
over seconds or even minutes to know, for example, if she is being asked for
information or ordered to do something.
Moreover, the processing of the sensory signal itself is only the
beginning, as the listener must search for familiar patterns in the input, such
as words and grammatical constructions, retrieve various sorts of information
associated with each, and combine them appropriately.
The crux of the problem,
then, is that language comes in sound by sound or fixation by fixation and yet
must be appreciated at word, sentence, and discourse levels. Part of what makes language processing so
difficult is that it requires so many levels of analysis and would seem to
recruit so many different subprocesses.
In addressing the question of how the human brain might approach this
task, many psycholinguists have sought refuge in the hierarchical structure of
language and assumed that the brain exploits this structure level by level,
using the outputs of low-level processes as inputs for the construction of
meaning at the sentence and discourse level.
On this approach, the role of the word is taken to be somewhat analogous
to the role of the cell in biology (e.g., Balota, 1994): although not the most
primitive level of analysis, it is assumed to be central to the comprehension
process.
Many classic models
include the assumption that word-level (lexical) processing necessarily
precedes processing at higher levels of analysis. Accordingly, psycholinguists
traditionally divide word comprehension into two stages, lexical and
post-lexical (e.g., Fodor, 1983; Forster, 1981; Seidenberg, Tanenhaus, Lieman,
& Bienkowski, 1982; Swinney, 1991).
Lexical processing is a fast, automatic stage of processing that
involves recognition, in which the phonemic or orthographic input is
matched to templates in the mental lexicon, and access, in which
syntactic and semantic information associated with the word is activated and
thereby made available to later stages of processing. In these later stages, known as post-lexical processing,
slower, controlled processes operate to combine linguistic information with
contextual and background knowledge in order to integrate the new word with the
developing representation at the message level.
Lexical Access
In the psychological
literature, the lexical access process has been studied by testing various
factors that affect performance on tasks such as naming and lexical decision, which
guarantee that the lexical entry has been accessed. This research has identified several fundamental phenomena as
characterizing the process. Of these,
the most fundamental is the frequency effect, the finding that less time is
needed to respond to high frequency words than to low frequency words (Balota
& Chumbley, 1984; Forster & Chambers, 1973; Taft, 1979; Whaley,
1978). Frequency effects have been
accounted for in two different, but related, ways. On models in which lexical access is construed as the result of
activation, frequency effects are thought to result because high frequency
words have a higher resting level of activation than low-frequency words (e.g.,
Morton, 1969), or because connections between the input and the lexical level
are weighted more heavily for high than low frequency words (e.g., Seidenberg
& McClelland, 1989). On models in
which lexical access is construed, instead, as the result of a search process
(e.g., Forster, 1979), it is assumed that words are stored and consequently
searched in the order of their frequency.
In either case, frequency facilitates the identification and access
processes so that lexical information is available more quickly for high
frequency words than for low.
Another watershed
finding is the semantic priming effect, or the finding that response latencies
to a word are reduced when it is preceded by a semantically or associatively
related word (Meyer & Schvaneveldt, 1971).
Though the details differ, a majority of the models of lexical access
explain semantic priming as resulting from the spread of activation or
excitation through the semantic network.
The reason people respond more quickly to "dog" when it
follows "cat" than when it follows "cloud" is that the
representations for the two related words are connected via an associative
pathway. Much work in the semantic
priming literature has assessed the automaticity of this process, where
automatic processes are defined as those which are fast-acting, and independent
of subjects' conscious control (Posner & Snyder, 1975).
A third finding is that
people are faster to respond to a word the second time it is presented than the
first, a phenomenon known as repetition priming (Forbach, Stanners, Hochhaus,
1974; Kirsner & Smith, 1974). On
activation models, repetition effects result because once a word's lexical
entry has been accessed, its activation decays slowly toward its resting
level. Similarly, on search models,
repetition effects result because the lexical entry remains "open"
for a short while after the access process (Forster & Davis, 1984). Forster and Davis (1984) have also proposed
that the initial presentation of a word results in the formation of an episodic
trace which facilitates the decision component of lexical decision, without
necessarily speeding lexical access.
Taken together, these
findings suggest that word information is organized associatively and that
access is affected by the frequency and recency with which that lexical entry
has been used in the past. Further, researchers
using the lexical decision task have identified two effects that suggest
important constraints on the architecture of the word identification
system. The first is the lexical status
effect, the finding that words elicit shorter response times than nonwords
(Rubenstein, Garfield, & Milikan, 1970).
The second is the nonword legality effect, the finding that pseudowords
with legal orthographic structure elicit slower response times than nonwords
constructed from random letter strings (Stanners, Forbach, & Headley, 1971;
Stanners & Forbach, 1973). These findings suggest that while the word
identification system can reject orthographically illegal strings out of hand,
pseudowords are treated differently and perhaps processed for some time as
words before being rejected.
Methodological Difficulties
As it happens,
interpretation of priming phenomena is not as straightforward as one would
like, and a certain amount of controversy surrounds almost all of the priming
effects described above. In each case
the controversy concerns whether the effect reflects facilitation in the
lexical stage of processing or in the post-lexical stage. This is particularly true of semantic
priming, and disputes over the automaticity of the processes underlying these
effects. Theoretically, these disputes
concern the issue of whether the associative organization suggested by priming
phenomena is a feature of the mental lexicon, or whether it reflects processes
involved in integration at other levels.
Methodologically, these
disputes have been rooted in claims about the influence of lexical versus
post-lexical processes on the experimental tasks used to elicit semantic
priming (reviewed in, e.g., Neely, 1991).
For example, reaction time paradigms are subject to criticism because
they require participants to terminate linguistic processing and to initiate
performance of the experimental task.
The imposition of a secondary task makes it much harder for the
psycholinguist to know what portion of the variance in observed priming effects
is attributable to lexical access, and what portion is caused by extra-lexical
processing related to task performance.
Uncovering information
about the specific nature and time course of lexical access is even more
difficult because one must make inferences from measures taken at the end point
of processing -- the vocal output or the button press -- rather from more
direct access to the intermediate stages of processing. Indeed, modern neuroimaging techniques are exciting
precisely because they promise to afford more direct access to these
intermediate stages. By measuring
oxygenation and blood flow associated with neural activation, positron emission
tomography (PET) and functional magnetic resonance imaging (fMRI), can indicate
which brain areas are active during particular experimental tasks. Similarities and differences in activation
patterns might then be used to help interpret discrepancies between different
behavioral tasks thought to index lexical access.
However, for all their
strengths, the temporal resolution of hemodynamic measures such as PET and fMRI
is fundamentally limited by the character and time course of the signal they
measure. Even the most optimistic
assessment of the temporal resolution of fMRI puts it between 0.5 and 1 second
(Friston, Fletcher, Josephs, Holmes, Rugg, and Turner, 1998), suggesting this
method is not adequate for probing the transient process of lexical
access. To directly answer questions
about the when, how, and what of lexical access, one must instead turn to techniques
that monitor the electrical activity of neurons in real time. Although the currents associated with the
firing of a single neuron are too small to be detected non-invasively, when a
large pool of neurons oriented in roughly the same direction fire
synchronously, their summed electrical activity is often detectable at the
scalp (Kutas & Dale, 1997). The
electrical fields that result when such pools of neurons -- particularly
pyramidal cells of the neocortex -- polarize or depolarize are measured in
electroencephalography (EEG), by placing electrodes on the surface of the
scalp.
Rather than analyzing
EEG elicited in the course of cognitive processing, electrophysiologists
interested in cognition have found it useful to examine event-related brain potentials
(ERPs). ERPs are patterned voltage
changes in the on-going EEG that are time locked to the onset of a
stimulus. The ERP is obtained by
recording subjects' EEG and then averaging across time-locked stimuli. This allows the researcher to focus on
aspects of the EEG that are temporally correlated with the processing of the
class of eliciting events. One
particular advantage of ERPs is that they can serve as an on-line measure of
subjects' processing that does not interfere with language processing itself.
Averaging the EEG time
locked to the presentation of words typically results in a series of positive
and negative peaks in the ERP known as components. Because these peaks are visually salient and
seem to be modulated in functionally interpretable ways, cognitive
electrophysiologists have devoted a good deal of attention to their
characterization. Components are
typically identified by their polarity, that is, whether they represent
positive or negative going voltage, and their latency with respect to the
eliciting event. For example, the P300 is a positive-going component that
typically peaks 300 ms after the onset of an unusual task-relevant event.
Besides polarity, latency, and functional characterization, an important aspect
of the definition of an ERP component is its topography, or amplitude
distribution over the scalp. To be
confident that one is looking at the "same" activity in two tasks,
for example, it is important that the distribution of the activity elicited in
a particular time-window be similar in both cases. Though the spatial resolution of ERPs is limited, differences in
the topography of two ERP effects almost certainly indicate differences in the
underlying neural generators of those effects.
Lexical and Post-Lexical Priming Effects
on the N400
Interestingly, cognitive
electrophysiologists have identified a component in the brainwaves that is
modulated by many of the same variables that affect tasks taken to index
lexical access. The N400 is a
negative-going wave evident between 200 and 500 ms after the presentation of a
word, so-called because it tends to peak around 400 ms after stimulus onset
(Kutas & Hillyard, 1980). Though
this effect is observed all over the scalp, it is largest over centroparietal
areas and is usually larger on the right side of the head than the left (Kutas,
Van Petten, & Besson, 1988). The
N400 is elicited by words in all modalities, whether written, spoken, or signed
(Holcomb & Neville, 1990; Kutas, 1987).
Moreover, the size, or amplitude, of the N400 is affected in a way that
is analogous in many respects to response times in naming and lexical decision
tasks.
For instance, in both
word lists and in sentences, high frequency words elicit smaller N400s than low
frequency words (Smith & Halgren,1989).
The N400 also evidences semantic priming effects, in that the N400 to a
word is smaller when it is preceded by a related word than when it is preceded
by an unrelated word (Bentin, 1985; Holcomb, 1988). Third, the N400 is sensitive to repetition -- smaller to
subsequent occurrences of a word than to the first (Rugg, 1985; Van Petten,
Kutas, Kluender, Mitchiner, & McIsaac, 1991). Furthermore, while pseudowords elicit even larger N400s than do
real words, orthographically illegal nonwords elicit no N400 at all (Kutas
& Hillyard, 1980).
At least at first
glance, then, the N400 seems like it might provide a window into lexical access
-- into a period of processing when pseudowords are still being distinguished
from real words and when variables like frequency, repetition, and semantic
relatedness have a significant impact.
However, in addition to its sensitivity to variables thought to
influence lexical access, the N400 is sensitive to factors that many would
consider post-lexical. For example, one
of the best predictors of N400 amplitude for a word in a given sentence is that
word's cloze probability (Kutas, Lindamood, & Hillyard, 1984). N400 amplitudes are largest for unexpected
items such as semantic anomalies, and smallest for contextually congruous words
with high cloze probabilities. In
general, N400 amplitude varies inversely with the predictability of the target
word in the preceding context.
To complicate matters
further, N400 is also sensitive to discourse-level factors such as the
activation of contextually relevant schemas.
For example, St. George, Mannes, & Hoffman (1994) recorded
participants' ERPs as they read ambiguous paragraphs that either were or were
not preceded by a disambiguating title.
Although the local contextual clues provided by the paragraphs were
identical in the titled and untitled conditions, words in the untitled
paragraphs elicited greater amplitude N400s.
Similarly, Van Berkum, Hagoort, & Brown (1999) found that words
which elicit N400s of approximately equal amplitude in an isolated sentence, do
not elicit equivalent N400s when they occur in a context that makes one version
more plausible than the other. For
instance, "quick" and "slow" elicit similar N400s in
"Jane told her brother that he was exceptionally quick/slow." However, "slow" elicits a much
larger N400 when this same sentence is preceded by "By five in the
morning, Jane's brother had already showered and had even gotten dressed."
This sensitivity of the
N400 to higher-order aspects of language would seem to rule it out as a valid
index of lexical access. However, eye
tracking and behavioral measures have also suggested that higher-order aspects
of language can have strong effects on the processing of individual words. For example, Duffy, Henderson, & Morris
(1989) measured naming latencies for words in sentence contexts and found that
both lexical and sentential factors affected priming. Moreover, sentence level
congruity elicited priming even when none of the individual words in the sentence
worked as associative primes.
Similarly, priming effects as indexed by phoneme monitoring and
cross-modal naming latencies also suggest that when words are presented in
contexts such as sentences and paragraphs, priming depends on congruity with
the larger context, rather than local associations between words in the
sentence (Foss & Speer, 1991; Hess, Foss, & Carroll, 1995). ERP data reviewed above are thus congruent
with these behavioral findings and, further, suggest that lexical, sentential,
and discourse factors all affect the brain response to words in the same time
window and in a qualitatively similar manner.
Context, Context, Context
When we look
independently at the effects of various kinds of linguistic variables on the
brain's activity, we find that they all manifest as a change in the amplitude
of the scalp-recorded N400. This is
surprising because, based on the models we have assumed thus far, we expect
differences in the time course, mechanism, and automaticity of the two stages
of processing. Lexical processing is
thought to be early, fast-acting, and automatic. Post-lexical processing is
thought to be slower, more controlled, and to require the output of the lexical
processing system for its operation.
Consequently, lexical variables are predicted: (1) to have an earlier
influence on word processing than post-lexical variables; (2) to influence word
processing via a qualitatively different mechanism than post-lexical variables;
and (3) to be more robust than post-lexical variables, as post-lexical
processing mechanisms depend on the completion of lexical processing.
In contrast to these
predictions, N400 priming effects seem to reflect constraints arising from
several different levels: lexical variables such as orthographic regularity,
frequency, repetition, and semantic relatedness, as well as post-lexical
variables such as predictability in the sentence context, congruity with the
overall discourse context, and activation of a contextually relevant schema. The similarity in the timing and the scalp
distribution of the various N400 priming effects do not bear out the prediction
that lexical processing occurs earlier, and employs a distinct processing
mechanism. However, the mere fact that
all these different variables influence N400 amplitude does not necessarily
imply that a single process subserves lexical, sentential and discourse level
processing. It could be the case for
instance, that lexical and post-lexical effects on the N400 are independently
generated. To assess this possibility,
it is necessary to compare the effect of lexical and post-lexical variables in
a single group of participants in cases where the task conditions are closely
matched.
Accordingly, Kutas
(1993) compared the effects of lexical and sentential context by comparing the
N400 priming effect in sentences and word pairs. During the first half of the experiment, participants' ERPs were
recorded as they read sentences for comprehension. Half of these sentences ended with a predictable response (e.g.,
"Before exercising Jake always stretches his muscles.") The other half ended with a congruous but
unexpected word, such as "Fred put the worm on the table." In the second half of the experiment,
participants' ERPs were recorded as they read word pairs derived from the
sentences, such as "biceps - muscles" and "hook -
table". Kutas (1993) reasoned that this design would enable comparison
of ERPs to the same word in roughly similar lexical and sentential contexts.
Because qualitatively
different cognitive and neural mechanisms should elicit qualitatively different
ERP effects, one can use the ERP measure to compare the time course and nature
of the mechanism that underlies word pair priming with those that underlie the
integration of word and sentence meaning.
However, Kutas (1993) found that the ERPs elicited by sentence final
words were qualitatively similar to those elicited by the second word in each
word pair. In fact, the priming effect
(i.e., the reduction in the N400 amplitude to the more predictable sentence
completion or the related word), was larger and began earlier in the sentence
condition than in the word pairs. Kutas
(1993) argued that the observed similarities in the waveshape, scalp
distribution, and experimental modulation of the N400 in lexical and sentential
contexts strongly suggest that a single process underlies both sorts of priming
effects.
In a similarly motivated
experiment, Van Petten (1993) examined the relationship between lexical and
post-lexical processing mechanisms by looking directly at the interaction of
semantic priming and sentence context effects within the same material. Van Petten manipulated sentential context by
comparing congruous sentences such as "When the moon is full it is
hard to see many stars of the Milky Way," with structurally similar
incongruous sentences such as "When the moon is rusted it is
available to buy many stars of the Santa Ana." Lexical context was manipulated by including
conditions with unrelated word pairs (such as "insurance" and
"refused"). For example,
"moon" and "stars" in the congruous and incongruous
sentences above could be compared with "insurance" and
"refused" in "When the insurance investigators found out that
he'd been drinking they refused to pay the claim" and "When the
insurance supplies explained that he'd been complaining they refused to speak
the keys." Like Kutas (1993), Van
Petten (1993) found that ERP effects elicited by the manipulation of lexical
and sentential context were strikingly similar in their polarity, waveshape, and
distribution over the head.
The fact that lexical
and post-lexical ERP effects display a similar time course argues against a
serial model in which lexical processing precedes post-lexical processing. Moreover, Van Petten and colleagues (1999)
present data that suggest that semantic integration can begin before word
recognition could possibly be complete. In an experiment designed to compare
the relative timing of phonemic analysis and sentential context effects, Van
Petten and colleagues (1999) collected ERPs to words in spoken sentences such
as, "The highway was under construction, so they had to take a lengthy
detour." Besides congruous
expected endings, Van Petten et al. included three types of unexpected endings:
plain incongruous, unexpected words (such as "table") with no
phonological similarities to the expected completion, rhyme incongruous,
(such as "contour") that shared the same final syllable as the
expected ending, and cohort incongruous, (such as "detail")
that shared the same initial syllable as the expected ending.
If semantic integration
requires the prior output of completed word recognition processes, we would
expect the N400 to be relatively insensitive to the phonemic similarity of the
various incongruent endings with the expected completion. However, if semantic integration can proceed
even on the basis of partial information, we might expect the N400 effect to
begin earlier in the plain incongruous words than in the cohort incongruous
words that share an initial syllable with the expected completion. Van Petten and colleagues (1999) found that
the N400 effect for plain and rhyme incongruous words began approximately 200
ms after the onset of the word, while the N400 effect for cohort incongruous
words like "detail" did not begin until 375 ms post-onset.
The later onset of the
N400 effect in the words that initially sounded like the expected completion
suggests that sentential integration can operate
on the basis of partial information.
However, definitive evidence on this issue requires knowledge of the
recognition point for each of the words, that is, the point in time at which
participants were able to
accurately recognize words based on partial phonemic input. Because Van Petten and colleagues (1999) had
previously established the recognition point for these words in a gating task,
they were able to record ERPs time locked to the recognition point for each
sentence-final word. In this
comparison, Van Petten and colleagues found that while ERPs to cohort
incongruous words like "detail" diverged at their recognition
point, ERPs to incongruous words like "table" and "contour"
diverged well before. This
finding indicates that semantic integration can proceed with partial phonemic
input, before word recognition is complete, and thus does not necessarily
require a prior stage of lexical access.
Evidence reviewed above
thus suggests that lexical and post-lexical variables operate similarly in a
similar time-window. Moreover, Van
Petten et al.’s findings suggest that the “postlexical” process of integration
can begin before lexical processing is fully complete. However, what about the final prediction,
that lexical effects are somehow more basic?
It might be the case, for example, that even if the stages are not as
discrete or mechanistically distinct as we thought, lexical effects are more
automatic, and less subject to outside influences.
When word and context collide
To test this latter
hypothesis, Van Petten and Kutas (1990) compared the effects of word frequency
and ordinal position in a sentence to see if the two factors were additive or
interactive. These researchers assumed
that frequency effects index lexical-level processing and ordinal position
effects index the increasing influence of context as the representation of sentence-level
meaning is built up. They reasoned
further that if the lexical-level processes that underlie frequency effects are
independent of the sentence-level processes that underlie ordinal position
effects, the two variables should influence the response measure autonomously
and have additive effects on N400 amplitude.
However, if the processes underlying frequency and ordinal position
effects are not independent of one another, the two factors should interact --
and the character of their interaction might be instructive about the
underlying mechanisms of priming.
In fact, Van Petten
& Kutas (1990) found main effects of frequency, position, and an
interaction between the two. Low
frequency words elicited larger N400
components than did high frequency words. Moreover, when Van Petten & Kutas
examined data from the sentence intermediate words on a closer basis they found
a downward linear trend in the N400 amplitude with word position. However, the frequency effect was most
prominent in words in the beginning of sentences, somewhat attenuated in
sentence-intermediate words, and entirely absent from sentence-final
words. Van Petten & Kutas argued
that this interaction, especially the fact that sentence-final words exhibit no
N400 frequency effect whatsoever, suggests the buildup of contextual knowledge
influences the initial activation and retrieval of lexical items, as well as
integrative processes typically referred to as post-lexical.
Other recent findings
also call into question the automaticity and primacy of so-called lexical level
variables. For example, Coulson and
colleagues (2000) compared ERPs to lexically associated words (such as
"life" and "death"), in contexts where they either made
sense together, as in, "When someone has a heart attack a few minutes can
make the difference between life and death," or did not, as in, "The
gory details of what he had done convinced everyone that he deserved life in
death." Although the word pairs
used in the study elicited N400 priming effects when presented in word lists,
they did not do so in these sentence contexts.
The finding that sentential context can apparently eliminate lexical
priming effects argues against the idea that this sort of priming reflects automatic
lexical access.
Camblin, Federmeier,
& Kutas (2000) compared N400 priming effects for lexical associates (and
matched nonassociates) that occurred within the same clause, as in "The
book described the kings and the queens but lacked details she
needed," with the same word pairs separated by a clause boundary, as in
"The book described the kings, but the queens were not
mentioned at all." Camblin et al.
found that the same pairs of words separated by the same number of words
generate ERP association effects with a different time course and distribution
when separated by
a clause boundary than when they
occur within the same clause. When the
associated words occurred in the same clause, the second word elicited a
canonical N400 priming effect. However,
when the words were separated by a clause boundary, the priming effect was
later and more frontal in distribution.
Taken together, these
findings suggest that the impact of lexical variables such as frequency and
lexical association is greater when higher-level contextual information is
impoverished. Frequency affects N400
amplitude more in the initial words of a sentence than at the end. When lexical association and contextual
congruity both help the reader to construct the message-level representation,
they each affect N400 amplitude (Van Petten, 1993). However, when lexical
association and contextual congruity conflict in their utility for the
construction of the message-level representation, contextual congruity
overrides the effect of lexical association.
In sum, putatively lexical effects assumed to be automatic and
independent seem to be modulated by higher-order aspects of context, either
semantic (as in Coulson et al., 2000) or syntactic (as in Camblin, et al.,
2000).
Moreover, these findings
in the ERP literature are buttressed by similar findings from psycholinguists
using more conventional methods. For example, Cutler, Norris, McQueen, and
Butterfield (2000), using a manipulation similar to that employed by Coulson et
al. (2000), found that cross-modal associative priming disappears in sentence
contexts. Murray (2000) has presented eye tracking data collected for sentences
where lexical associations are pitted against grammatical cues. He reports clear effects of structural
factors, but no lexical priming, echoing Camblin et al.'s finding that
grammatical factors affect lexical priming.
Finally, Traxler and colleagues (submitted) have presented a series of
eye tracking studies designed to uncover the nature of lexical priming effects. They crossed a variety of lexical-level
variables with plausibility, and found that, while plausibility always affected
reading times, among the lexical variables only repetition affected early
processing measures, such as first fixation and gaze duration. Traxler and colleagues argue that their data
undermine models that advocate an important role for intralexical spreading
activation, and instead suggest that priming effects reflect facilitated
integration into the discourse model.
Conclusions
We began with a
reasonable assumption, seen often in the psycholinguistic literature, that
language is constructed hierarchically, with the word as the basic unit of
analysis. On such a view, a very basic,
perhaps automatic, lexical access process is needed to identify words and call
up word-related information. Only after
lexical access has taken place can higher order semantic and
syntactic representations be built
up, via a variety of post-lexical mechanisms linking linguistic and
nonlinguistic information.
However, the data
reviewed above call into question the assumption that there are mechanistically
and temporally distinct stages for "lexical access" and
"post-lexical" context-based, integrative effects. The amplitude of
the N400 component has been modulated by variables such as frequency,
repetition, associative priming, lexicality, and orthographic legality, that
are known to influence word recognition. Moreover, N400 amplitude is also
modulated by variables, such as predictability in the sentence, that are
thought to influence post-lexical integration, as well as by variables that
influence still higher levels of comprehension, such as congruency with the
larger discourse context. Further, when
lexical (e.g. frequency, association) and post-lexical (e.g. congruency)
variables have been jointly manipulated, they both affect the ERPs beginning
around 200 ms after stimulus onset and seem to similarly modulate the neural
generators whose activity is visible at the scalp as the N400 effect.
The fact that activity
in the same brain areas is modulated by lexical and post-lexical variables is
not consistent with models that posit discrete mechanisms for lexical and
post-lexical priming effects. Further, the similarity in the time course of the
effects elicited by manipulations of lexical and post-lexical variables
indicates that the underlying process or processes operate at nearly the same
time. Moreover, the serial model, in
which lexical processing necessarily precedes post-lexical contextual
integration is undermined by Van Petten et al.'s (1999) finding that the N400
effect can begin before word recognition is even complete. Far from being impervious to outside
influence, lexical effects can be modulated by contextual factors, and can even
be eliminated. Such findings argue
against an underlying process that is automatic.
Moreover, while most
disputes about whether a particular variable operates lexically or
post-lexically are rooted in claims about the influence of these processes on
the experimental task, findings reviewed here concern electrophysiological
measures collected without the imposition of potentially confounding task
demands. Consequently, N400 priming effects cannot be dismissed merely because
they are post-lexical. While so-called
post-lexical influences on, for instance, the lexical decision task, may be
presumed irrelevant to normal language comprehension, the effects of high- as
well as low- level information on N400 amplitude are a natural consequence of
the word processing mechanisms related to its elicitation. Although effects of frequency, repetition,
and lexical associates have traditionally been considered relevant to the
initial stages of word recognition, data reviewed above suggest these factors
are actually relevant to the integration of word meaning with contextual
information (see also Traxler et al., submitted).
In
other words, the interaction of low-level variables such as word association
and frequency with higher-level manipulations of context suggests the brain
processes that subserve language comprehension do not make the same
distinctions as the psycholinguist.
Rather, the brain seems to use all the information that it can as early
as it can to constrain the comprehension task, from the analysis of the
perceptual signal through word recognition and the construction of pragmatic
and discourse models. This way of
thinking about language processing mirrors in some ways a recent shift in
theories of visual scene processing. In
these theories, it was once thought that the process of understanding a scene
was completely hierarchical, involving an initial stage in which the raw
sensory input was analyzed so that objects could be identified, and a
subsequent set of post-identification processes that computed the relationships
between identified objects in order to construct the scene. However, the growing consensus is that
processing actually takes place at multiple levels of analysis in parallel,
such that object identification and scene analysis are to some extent
interdependent (e.g., review in Henderson & Hollingworth, 1999). On the modern view of high level visual
processing, object recognition and scene analysis, though separable, are
interactive and mutually constraining.
In
the end, a scene is more than a collection of objects, just as a sentence is
more than a collection of words. While
processing language surely involves retrieving information associated with
incoming words, it also requires the integration of lexical information with
higher-order grammatical representations, with local, contextual information,
and with the background knowledge activated by discourse-level representations
to fully understand what those words convey (Coulson, in press). For example, to understand the significance
of “water” in “Everyone had so much fun diving from the tree into the pool, we
decided to put in a little water,” it is not sufficient to know just that
“water” is a mass noun, and that it represents a colorless, odorless liquid
essential for plant and animal life. To
integrate “water” into the context and appreciate the joke, it is necessary to
exploit grammatical knowledge (i.e. gapping) to infer that the water goes into
the pool, contextual knowledge to know that the agents are diving from the tree
into a pool that either does or does not have water in it, and background knowledge
to know that pools without water are a lot less fun to dive into. Given the complexity of contextual
integration and the necessity of rapidly combining information from multiple
levels of analysis, the brain’s approach – concurrent utilization of
information about orthography, recency, frequency, lexical association, and
congruity at the sentence- and discourse- levels – seems well-suited for the
task of constructing the message. In
sum, emerging data suggest the brain does not process words and context,
but words in context.
References
Balota,
D. (1994). Visual word recognition. In
M. Gernsbacher (Ed.), Handbook of Psycholinguistics. New York: Academic Press, Inc., pp. 303-358.
Balota,
D.A. & Chumbley, J.I. (1984). Are lexical decisions a good measure of
lexical access? The role of word-frequency in the neglected decision
stage. Journal of Experimental
Psychology: Human Perception and Performance 10, 340-357.
Bentin,
S., McCarthy, G., & Wood, C. C. (1985). Event-related potentials associated
with semantic priming. Electroencephalography
and Clinical Neurophysiology, 60, 343-355.
Camblin,
C.C., Federmeier, K.D., & Kutas, M. (2000, March). The effect of clause boundaries on the
processing of related words: An electrophysiological analysis. Poster presented at the 13th
Annual CUNY Conference, San Diego, CA.
Coulson,
S. (2000, in press). Semantic Leaps: Frame-shifting and Conceptual Blending
in Meaning Construction. Cambridge
and New York: Cambridge University Press.
Coulson,
S., Van Petten, C., Federmeier, K., Folstein, J., Weckerly, J., & Kutas, M.
(2000, March). Lexical and sentential
context effects: An ERP study of the difference between life and death and life
in prison. Paper presented to the 13th
Annual CUNY Conference, San Diego, CA.
Cutler,
A., Norris, D., McQueen, J., & Butterfield, S. (2000, March). Cross-modal
associative priming which disappears in sentence context. Poster presented at the 13th
Annual CUNY Conference, San Diego, CA.
Duffy,
S. A., Henderson, J. M., & Morris, R. K. (1989). Semantic facilitation of
lexical access during sentence processing. Journal of Experimental
Psychology: Learning, Memory, & Cognition, 15(5), 791-801.
Fodor,
J. (1983). The Modularity of Mind.
Cambridge, MA: MIT Press.
Forbach,
G., Stanners, R., & Hochhaus, L. (1974). Repetition and practice effects in
a lexical decision task. Memory & Cognition 2, 337-339.
Forster,
K.I. (1979). Levels of processing and the structure of the languagae
processor. In W.E. Cooper & E.
Walker (Eds.), Sentence processing.
Hillsdale, NJ: Erlbaum Associates Limited, pp. 27-85.
Forster,
K.I. (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.
Forster,
K.I. & Chambers, S.M. (1973). Lexical access and naming time. Journal of
Verbal Learning and Verbal Behavior 12, 627-635.
Forster,
K.I. & Davis, C. (1984). Repetition priming and frequency attenuation in
lexical access. Journal of
Experimental Psychology: Learning, Memory, and Cognition 10, 680-698.
Foss,
D.J. & Speer, S.R. (1991). Global and local context effects in sentence
processing. In R.R. Hoffman & D.S.
Palermo (Eds.), Cognition and Symbolic Processes: Applied and Ecological
Perspectives, pp. 115-139.
Hillsdale, NJ: Erlbaum.
Hess,
D.J., Foss, D.J. & Carroll, P.
(1995). Effects of local and global context on lexical processing during
language comprehension. Journal of
Experimental Psychology: General 124, 62-82.
Holcomb,
P. J. (1988). Automatic and attentional processing: An event-related brain
potential analysis of semantic priming. Brain & Language, 35(1),
66-85.
Holcomb,
P. J., & Neville, H. J. (1990). Auditory and visual semantic priming in
lexical decision: A comparison using event-related brain potentials. Language
& Cognitive Processes, 5(4), 281-312.
Kirsner,
K. & Smith, M.C. (1974). Modality effects in word recognition. Memory
& Cognition 2, 637-640.
Kutas,
M. (1993). In the company of other words: Electrophysiological evidence for
single-word and sentence context effects. Special Issue: Event-related brain
potentials in the study of language. Language & Cognitive Processes, 8(4),
533-572.
Kutas,
M., & Dale, A. (1997). Electrical and magnetic readings of mental
functions. In M. D. Rugg (Ed.), Cognitive Neuroscience (pp. 197-242).
Hove, East Sussex: Psychology Press.
Kutas,
M., & Hillyard, S. A. (1980). Reading senseless sentences: Brain potentials
reflect semantic incongruity. Science, 207(4427), 203-205.
Kutas,
M., Lindamood, T. E., & Hillyard, S. A. (1984). Word expectancy and event-related
brain potentials during sentence processing. In S. Kornblum & J. Requin
(Eds.), Preparatory States and Processes (pp. 217-237). Hillsdale, NJ:
Lawrence Erlbaum Associates.
Kutas,
M., Neville, H. J., & Holcomb, P. J. (1987). A preliminary comparison of
the N400 response to semantic anomalies during reading, listening, and signing.
Electroencephalography and Clinical Neurophysiology, Supplement 39,
325-330.
Kutas,
M., Van Petten, C., & Besson, M. (1988). Event-related potential
asymmetries during the reading of sentences. Electroencephalography &
Clinical Neurophysiology, 69(3), 218-233.
Meyer,
D. & Schvaneveldt, R. (1971).
Facilitation in recognizing pairs of words: Evidence of a dependence
between retrieval operations. Journal
of Experimental Psychology 90, 227-234.
Morton,
J. (1969). Interaction of information
in word recognition. Psychological
Review 76, 165-178.
Murray,
W.S. (2000, March). Meaning, structure, and the interpretive process. Poster presented at the 13th
Annual CUNY Conference, San Diego.
Neely,
J. H. (1991). Semantic priming effects in visual word recognition: A selective
review of current findings and theories. In D. Besner & G. W. Humphreys
(Eds.), Basic processes in reading: Visual word recognition. (pp.
264-336). Hillsdale, NJ: Lawrence Erlbaum Associates.
Posner,
M.I. & Snyder, C.R. (1975). Attention and cognitive control. In R.L. Solso
(Ed.), Information Processing and Cognition: The Loyola Symposium.
Hillsdale, NJ: Lawrence Erlbaum Associates, Inc, pp. 55-85.
Rubenstein,
H., Garfield, L., & Milikan, J.A. (1970). Homographic entries in the
internal lexicon. Journal of Verbal Learning and Verbal Behavior 18,
757-767.
Rugg,
M. D. (1985). The effects of semantic priming and word repetition on
event-related potentials. Psychophysiology, 22(6), 642-647.
Seidenberg,
M.S. & McClelland, J.L. (1989). A distributed, developmental model of word
recognition and naming. Psychological Review 96, 523-568.
Seidenberg,
M.S., Tanenhaus, M.K., Leiman, J.M., & Bienkowski, M. (1982). Automatic access of the meanings of
ambiguous words in context: Some limitations of knowledge-based
processing. Cognitive Psychology
14, 489-537.
Smith,
M.E. & Halgren, E. (1989). Dissociation of recognition memory components
following temporal lobe lesions. Journal of Experimental Psychology:
Learning, Memory, and Cognition 15, 50-60.
Stanners,
R.F. & Forbach, G.B. (1973). Analysis of letter strings in word
recognition. Journal of Experimental
Psychology 98, 31-35.
Stanners,
R.F., Forbach, G.B., & Headley, D.B. (1971). Decision and search processes
in word-nonword classification. Journal of Experimental Psychology 90,
45-50.
Swinney,
D. (1991). The resolution of
indeterminacy during language comprehension: Perspectives on modularity in
lexical, structural and pragmatic processing.
In G.B. Simpson (Ed.), Understanding Word and Sentence. Advances in Psychology, Vol 77. New York: North-Holland. Pp. 367-385.
Taft,
M. (1979). Recognition of affixed words and the word frequency effect. Memory
& Cognition 7, 263-272.
Traxler,
M.J., Foss, D.J., Seely, R.E., Kaup, B., & Morris, R.K. (2000, submitted).
Priming in sentence processing: Intralexical spreading activation, schemas, and
situation models.
van
Berkum, J. J. A., Hagoort, P., & Brown, C. M. (1999). Semantic integration
in sentences and discourse: Evidence from the N400. Journal of Cognitive
Neuroscience, 11(6), 657-671.
Van
Petten, C. (1993). A comparison of lexical and sentence-level context effects
in event-related potentials. Special Issue: Event-related brain potentials in
the study of language. Language & Cognitive Processes, 8(4),
485-531.
Van
Petten, C., Coulson, S., Rubin, S., Plante, E., & Parks, M. (1999). Time
course of word identification and semantic integration in spoken language. Journal
of Experimental Psychology: Learning, Memory, & Cognition, 25(2),
394-417.
Van
Petten, C., & Kutas, M. (1990). Interactions between sentence context and
word frequency in event-related brain potentials. Memory & Cognition, 18(4),
380-393.
Van
Petten, C., Kutas, M., Kluender, R., Mitchiner, M., & McIsaac, H. (1991).
Fractionating the word repetition effect with event-related potentials. Journal
of Cognitive Neuroscience, 3(2), 131-150.
Whaley,
C.P. (1978). Word-nonword classification time. Journal of Verbal Learning
and Verbal Behavior 17, 143-154.