ArticlePDF Available

Coreference and Lexical Repetition: Mechanisms of Discourse Integration

Authors:

Abstract and Figures

The use of repeated expressions to establish coreference allows an investigation of the relationship between basic processes of word recognition and higher level language processes that involve the integration of information into a discourse model. In two experiments on reading, we used eye tracking and event-related potentials to examine whether repeated expressions that are coreferential within a local discourse context show the kind of repetition priming that is shown in lists of words. In both experiments, the effects of lexical repetition were modulated by the effects of local discourse context that arose from manipulations of the linguistic prominence of the antecedent of a coreferentially repeated name. These results are interpreted within the context of discourse prominence theory, which suggests that processes of coreferential interpretation interact with basic mechanisms of memory integration during the construction of a model of discourse.
Content may be subject to copyright.
There are good reasons to expect that readers and lis-
teners process words differently when they are embedded
in a sentence than when those same words are embedded
in a word list that lacks linguistic structure. The goal of
sentence processing is to extract an integrated, coherent
message from the linguistic input, a goal that is absent
during the comprehension of the words in a list. In this
article, we examine the possibility that processing words
within a sentential context might result in effects other
than, and perhaps counter to, those observed during basic
lexical processing outside of a structured context.
Specifically, we address the behavioral and electro-
physiological consequences of encountering lexical rep-
etition in sentences. Studying the effects of repetition on
lexical processing has proven to be a remarkably fruitful
endeavor in cognitive psychology and cognitive neuro-
science, providing a great deal of information about the
retrieval of lexical information at different levels of pro-
cessing. A fundamental effect of lexical repetition, one
that plays an important role in research on word recogni-
tion and implicit memory, is repetition priming, a facilita-
tion in the processing of a word when that word has been
encountered previously.
The vast majority of studies on repetition priming
have examined the processing of words presented in lists,
which has been appropriate to the goals of those studies
(see, e.g., Jacoby & Dallas, 1981; Mimura, Verfaellie, &
Milberg, 1997; Scarborough, Cortese, & Scarborough,
1977). Here, our goal is to determine whether repetition
priming can be dissociated from effects that arise due
to sentential context. We used eye tracking and event-
related potentials (ERPs) during reading to examine how
the structure and meaning of a sentence modulate the ef-
fects of repeating a name. We did so in order to test the
hypothesis that in specific linguistic contexts, a repeated
name (the function of which is ultimately to establish
coreference) is, at some level, processed in the same way
as a novel name. This hypothesis is derived from a model
of discourse processing (Gordon & Hendrick, 1998b)
that addresses the establishment of coreference for dif-
ferent types of noun phrases (NPs) within and between
sentences.
Two linguistic expressions are said to be coreferential if
they refer to the same semantic entity; the first expression
(the antecedent) introduces the entity into the discourse
model, and the second expression (the anaphor) refers to
it. Coreference can occur both within and between sen-
tences and can be established by the use of full expres-
sions (such as descriptions or names), as well as reduced
expressions (such as pronouns and ellipses); it is a funda-
mental mechanism for making language coherent (Grosz,
Joshi, & Weinstein, 1995; Kintsch & van Dijk, 1978).
801 Copyright 2007 Psychonomic Society, Inc.
Coreference and lexical repetition:
Mechanisms of discourse integration
Kerry Ledoux
University of California, Davis, California
Peter C. Gordon and C. Christine CambLin
University of North Carolina, Chapel Hill, North Carolina
and
tamara y. swaab
University of California, Davis, California
The use of repeated expressions to establish coreference allows an investigation of the relationship between
basic processes of word recognition and higher level language processes that involve the integration of informa-
tion into a discourse model. In two experiments on reading, we used eye tracking and event-related potentials
to examine whether repeated expressions that are coreferential within a local discourse context show the kind
of repetition priming that is shown in lists of words. In both experiments, the effects of lexical repetition were
modulated by the effects of local discourse context that arose from manipulations of the linguistic prominence
of the antecedent of a coreferentially repeated name. These results are interpreted within the context of discourse
prominence theory, which suggests that processes of coreferential interpretation interact with basic mechanisms
of memory integration during the construction of a model of discourse.
Memory & Cognition
2007, 35 (4), 801-815
K. Ledoux, kledoux1@jhmi.edu
802 Ledoux, Gordon, CambLin, and Swaab
Theoretical analyses of coreference within the binding
theory (Chomsky, 1981) have focused on the interaction
between the syntactic position of the antecedent and the
form of the anaphoric expression. Of particular relevance
to this article, those analyses have explored conditions in
which two expressions exhibit disjoint reference—that is,
they cannot refer to the same thing. One condition under
which disjoint reference is purported to arise is when
the antecedent expression has a certain kind of syntactic
prominence
1
in relation to an anaphoric expression that is
a full NP, such as a name or a description. According to the
binding theory analysis, the two occurrences of “John” in
Example 1a cannot possibly refer to the same person; the
syntactic prominence relation of the antecedent to the NP
anaphor results in disjoint reference. On the other hand,
according to the binding theory it is possible (although not
necessary) that the two instances of “John” in Example 1b
do refer to the same entity, because the embedding of the
antecedent within the conjoined NP eliminates this syn-
tactic prominence relation:
1. a. John went to the store so that John could buy
some candy.
b. John and Mary went to the store so that John
could buy some candy.
These principles of binding theory were developed on
the basis of the metalinguistic judgments of linguists, not
on the basis of the judgments of individuals who were
naive to linguistic theory. Research using naive partici-
pants has yielded judgments of the acceptability of core-
ference that diverge sharply from those underlying the
binding theory (Gordon & Hendrick, 1997, 1998a; Keller
& Asudeh, 2001). Gordon and Hendrick (1997) found that
embedding an antecedent name in a conjoined NP (i.e., the
contrast between 1a and 1b) increased the proportion of
responses that deemed coreference acceptable with a sub-
sequent repeated name. However, this effect was small, as
compared with that of other factors that influenced meta-
linguistic judgments of coreferential acceptability. In par-
ticular, coreference in name–name sequences was deemed
less acceptable than that in name–pronoun sequences, al-
though it was deemed far more acceptable than coreference
in pronoun–name sequences. This shows that pronouns
provide the most natural way of establishing coreference
with a previous name (at least for cases of within-sentence
coreference in which the pronoun is unambiguous) but
that coreference with repeated names is acceptable even
when the stimulus set offers an implicit comparison with
a coreferential configuration (name–pronoun sequences)
that is clearly better. Gordon and Hendrick (1997) also
found that the characteristics of prominence did not match
those described in the binding theory (Chomsky, 1981),
further indicating that characterizations of grammatical-
ity should not rest solely on the binding theory. Finally,
Gordon, Hendrick, Ledoux, and Yang (1999) reported
reading-time evidence showing a strong parallel in the
way in which the syntactic prominence of an anteced-
ent affected coreferential processing of repeated names
within and between sentences, a finding that suggests that
restricting the analyses of these effects to the domain of a
sentence (as in binding theory) misses important general-
izations about coreferential processes.
The theoretical framework of Gordon and Hendrick
(1998b), called discourse prominence theory, provides a
synthesis of these and other findings, using formalisms
adapted from model-theoretic semantics (Kamp & Reyle,
1993). It outlines procedures for incremental processing
of referential and coreferential NPs during the construc-
tion of a model of discourse and attempts to account for
the interplay of syntactic prominence with different types
of NPs, using the same interpretive mechanisms for core-
ference within sentences and between sentences in a local
discourse segment. Discourse prominence theory shares
basic features with other approaches that situate corefer-
ential processing during language comprehension within
the building of a model of discourse (e.g., Garnham, 2001;
Johnson-Laird, 1983; Sanford & Garrod, 1981). It differs
from other approaches that have emphasized semantic
factors, such as the scenario-mapping theory (Sanford &
Garrod, 1998; Sanford & Moxey, 1995), in that it has been
developed primarily to account for structural factors in
coreferential processing, particularly the connection be-
tween coreferential processing and grammatical theory.
Discourse prominence theory (Gordon & Hendrick,
1998b) specifies construction rules for interpreting dif-
ferent types of referential NPs in relation to the current
state of the developing discourse model. The construction
rule for interpreting names (and other full NPs) introduces
a new semantic entity to the discourse model on which
the name is predicated. In the case of a repeated name,
this will result in the discourse models containing two
distinct entities with the same name, providing the basis
for the intuitive sense of disjoint reference. Coreference
in such cases is achieved only by additional construc-
tion rules that operate to simplify the discourse model by
establishing equivalence between entities that match on
their predication. The magnitude of the sense of disjoint
reference and the difficulty of establishing coreference
between the repeated names is determined by the char-
acteristics of the antecedent expression. When the first
occurrence of the name has prominence in the discourse
model (as determined by syntactic factors related to the
height of the antecedent NP in the existing syntactic struc-
ture of the sentence and, perhaps, by semantic factors),
the existence of two distinct entities with the same name
is subjectively apparent and impedes the establishment
of coreference. Thus, a repeated name in this situation is
processed for an initial period of time as if it were a new
name. The additional processing that is needed to equate
the two instances of the repeated name results in a re-
peated name penalty when the antecedent is prominent
in the discourse representation (Almor, 1999; Garrod,
Freudenthal, & Boyle, 1994; Gordon, Grosz, & Gilliom,
1993; Gordon et al., 1999; Kennison & Gordon, 1997;
Yang, Gordon, Hendrick, & Hue, 2003). It is this predic-
tion from discourse prominence theory that we tested by
examining whether repetition priming would occur when
the antecedent was linguistically prominent.
In two experiments (the first using eye tracking,
and the second using ERPs), we used sentences such
CoreferenCe and LexiCaL repetition 803
as those presented in Example 2, in which the promi-
nence of the antecedent is manipulated by whether
or not it is embedded in a conjoined NP. A number of
studies have shown that this type of embedding af-
fects the prominence of an antecedent (Albrecht &
Clifton, 1998; Gordon & Hendrick, 1997, 1998b; Gordon
et al., 1999), and the manipulation has been successfully
used to study such important issues as the nature of the
representation of plural entities and the processes in-
volved in splitting a conjoined NP with singular reference
(Albrecht & Clifton, 1998; Carreiras, 1997; Kaup, Kelter,
& Habel, 2002; Koh & Clifton, 2002; Moxey, Sanford,
Sturt, & Morrow, 2004; Sanford & Lockhart, 1990).
2. a. Prominent NP1/repeated name
At the office Daniel moved the cabinet because
Daniel needed room for the desk.
b. Prominent NP1/new name
At the office Daniel moved the cabinet because
Robert needed room for the desk.
c. Nonprominent NP1/repeated name
At the office Daniel and Amanda moved the
cabinet because Daniel needed room for the
desk.
d. Nonprominent NP1/new name
At the office Daniel and Amanda moved the
cabinet because Robert needed room for the
desk.
In these sentences, prominence of the first name in the
sentence (the first instance of “Daniel” in the example)
was manipulated by whether it was the sentential subject
in a singular first NP (the prominent condition) or was
embedded in a sentential subject consisting of a con-
joined NP (the nonprominent condition); prominence is
inversely related to depth of syntactic embedding (Gordon
& Hendrick, 1998b). The subject of the second clause was
realized as a repeated name matching the first name in
NP1 (the second “Daniel” in 2a and 2c) or as a new name
(“Robert” in 2b and 2d).
If repetition effects are immune to the influence of
higher level processing, we might expect to see evidence
of repetition priming regardless of the structural con-
straints of the sentence. We would thus expect to see a
processing benefit for a repeated word, relative to a new
word, in conditions that are otherwise identical (in 2a vs.
2b and in 2c vs. 2d). We do not, however, expect this to be
the case; we expect, instead, that readers will be subject
to the influence of sentential context. Discourse promi-
nence theory, and the judgment and reading time data
collected in support of it, suggest a specific mechanism
for this influence in sentences such as those presented in
Example 2. In studies looking at explicit judgments of the
acceptability of coreference with repeated names, Gordon
and Hendrick (1997) showed that naive subjects consider
repeated-name coreference less acceptable when NP1 is
the sentential subject of a sentence (as in 2a) than when
it is embedded in a sentential subject consisting of a con-
joined NP (as in 2c); Gordon et al. (1999) showed parallel
results with self-paced reading. Coreference will be easily
and readily established using repeated names when the
antecedent is not prominent in the discourse model (2c);
in such sentences, the benefits of repetition and ease of
integration will conspire to facilitate processing. When
the antecedent is more prominent in the discourse model
(2a), repeated-name coreference will be difficult. The fa-
cilitation of processing that is conferred due to repetition
will be countered by a relative difficulty of integration due
to structural constraints of the sentence. We might, then,
predict a modulation of repetition priming in conditions
in which discourse prominence inhibits the establishment
of coreference.
As methods for studying online language comprehen-
sion, eye tracking and ERPs have complementary meth-
odological strengths: Eye tracking involves the normal
presentation of language stimuli and allows the speed and
location of linguistic information acquisition to proceed
naturally, whereas ERPs have distinctive components that
have been associated with different kinds of language
processes (Gordon, Camblin, & Swaab, 2004; Kutas &
Federmeier, 1998; Sereno & Rayner, 2003). To the extent
that our manipulations have strong effects on language
processing, we expected that eye-tracking and ERP re-
sults should converge. In particular, we predicted that dis-
course context (operationalized as the prominence of the
first NP) would modulate the effects of repetition priming
that might arise in the repeated name conditions. How-
ever, these methods differ in the timing in which stimulus
information is available and may be differentially sensi-
tive to some effects, as has been demonstrated by prior re-
search in which only one of the two methods (described in
more detail below) has been used. The use of both thereby
provides the opportunity to obtain a more complete view
of coreferential interpretation and repetition priming than
could be obtained with either method alone.
EXPERIMENT 1
In our first experiment, we used eye tracking during
reading to examine the factors of discourse prominence
and repetition. Previous research in which eye track-
ing has been used during reading (Garrod et al., 1994;
Gordon et al., 2004; Kennison & Gordon, 1997) has pro-
vided information on how the prominence of a discourse
referent influences subsequent coreferential interpreta-
tion of names and pronouns; however, that research did
not include a lexically matched baseline, such as the new-
name condition used here, for assessing the presence of
repetition priming. The present experiment tested whether
repeated-name coreference to a prominent discourse ref-
erent would modulate repetition priming effects in eye
tracking.
Lexical Repetition and Eye Movements
During Reading
Surprisingly, given the large amount of research on lex-
ical repetition in which a variety of behavioral-dependent
measures has been used, relatively little research has been
done on the topic with patterns of eye movements as the
dependent measure. Most of the research that has been
804 Ledoux, Gordon, CambLin, and Swaab
done has focused not necessarily on what happens when a
word is encountered a second time in a given text, but in-
stead on what happens when the same text is read a second
time, sometimes with alterations of specific words, in ef-
forts to understand the mechanisms that cause a text to be
read more quickly the second time (the rereading benefit,
or text repetition effect). Our experimental sentences, and
their use of lexical repetition, created processing demands
that were different from those created by the repetition of
an entire text. First, in text repetition research, the relevant
integration has been between the repeated word and ear-
lier portions of the repeated text, not between the repeated
word and the earlier instance of the repeated word, as it
was here. Also, in text repetition research, the critical re-
peated words have had greater separation than was the
case in our sentential stimuli, and those words have been
surrounded by other repeated words.
However, at least one finding from text repetition stud-
ies parallels that of eye-tracking studies that focus more
specifically on lexical repetition: Although both lexical
repetition and synonymy have been shown to influence
later eye-tracking measures associated with processes of
integration, often it is only lexical repetition that influ-
ences early eye fixation measures that are typically associ-
ated with processes of lexical access (Raney, Therriault,
& Minkoff, 2000). For example, in Raney et al., lexical
access alone was taken to be indicated by fixation dura-
tion when a word was fixated only once, whereas integra-
tion plus lexical access were taken to be indicated by the
sum of fixation durations when a word was fixated more
than once. Exact lexical repetition led to shorter fixation
times for instances of both single fixations and multiple
fixations, whereas synonym repetition led to shorter times
only in cases of multiple fixations.
A handful of studies have used eye tracking to examine
the modulation of lexical repetition effects by sentential
context. In their second experiment, Traxler, Foss, Seely,
Kaup, and Morris (2000) manipulated lexical repetition
and sentence plausibility in sentences such as those in Ex-
ample 3:
3. a. The lumberjack greeted the lumberjack early
this morning.
b. The young man greeted the lumberjack early
this morning.
c. The lumberjack chopped the lumberjack early
this morning.
d. The young man chopped the lumberjack early
this morning.
In 3a and 3c, the critical word (italicized in the example) is
a repetition of the sentential subject; in 3b and 3d, it is not.
In 3a and 3b, the critical word, when integrated with the
preceding sentential context, creates a plausible continua-
tion; in 3c and 3d, this word renders the sentence implau-
sible. The effects of these two manipulations were dissoci-
ated in the eye-tracking measures; Traxler et al. reported
a main effect of repetition (shorter times for repeated than
for new critical words) for first-fixation duration and gaze
duration but a main effect of plausibility for total read-
ing time. In other words, early measures were sensitive to
repetition priming, whereas later measures were sensitive
to sentential context (plausibility).
Liversedge, Pickering, Clayes, and Branigan (2003)
measured eye movements during the reading of adjunct
phrases for which thematic role assignment (temporal
or locative) was temporarily ambiguous. As is shown in
Example 4, target sentences (containing the ambiguous
adjunct) were preceded by a context sentence that biased
a temporal or a locative thematic role assignment:
4. a. Locative/locative
Context: The maid thought about where to peel/
prepare the vegetables.
Target: In fact, she peeled them in the kitchen,
with great care.
b. Locative/temporal
Context: The maid thought about where to peel/
prepare the vegetables.
Target: In fact, she peeled them in the morning,
with great care.
c. Temporal/locative
Context: The maid thought about when to peel/
prepare the vegetables.
Target: In fact, she peeled them in the kitchen,
with great care.
d. Temporal/temporal
Context: The maid thought about when to peel/
prepare the vegetables.
Target: In fact, she peeled them in the morning,
with great care.
Repetition of the verb across context and target sentences
was also manipulated (to differentiate between a lexical
locus and a discourse locus of the thematic biasing effect).
A main effect of verb repetition was found for first-pass
reading time of the region following the verb region (which
did not include the disambiguating noun; in the in the ex-
ample). Congruency effects (shorter times for congruent
contexts than for incongruent ones) were found for the two
regions following the disambiguating noun (the spillover
region [with great] and the final region [care]). Although
there was some indication that congruency influenced early
processing measures for these regions (with marginally
significant effects on first-pass reading time), the stron-
gest (statistically significant) effects were on the aggregate
measure of total reading time and on the later measure of
rereading. These findings again suggest a dissociation be-
tween repetition priming effects (which tend to influence
early eye-tracking measures) and context effects (which
have a greater effect on later measures). In this experiment,
as in others, these effects tended to occur downstream from
the critical words themselves, in spillover regions.
Our experiment, like the two mentioned above, involved
the manipulation of a lexical factor (repetition) and a con-
text factor (coreference with a prominent antecedent). On
the basis of previous results, we expected that these two
factors might influence different eye-tracking measures.
Early measures were expected to show evidence of repeti-
CoreferenCe and LexiCaL repetition 805
tion priming; that is, repeated names would be read more
quickly than new names when assessed by early measures
of processing. On the other hand, our theoretical analysis
of coreference focused on processes of integration of a
word’s meaning (specifically, its referent) into a discourse
model. In particular, integration of a repeated name with
a nonprominent antecedent was expected to be easier than
integration of a repeated name with a prominent anteced-
ent, relative to the baseline provided by the new, noncore-
ferential names. This ease of integration should result in
shorter reading times in eye movement measures associ-
ated with integration—that is, later processing measures.
These effects might be evidenced during the reading of
the critical word itself or might be displaced to regions
downstream in the sentence.
Method
Participants
. Forty students at the University of North Carolina
at Chapel Hill served as participants as part of a course requirement.
One participant was dropped from the analyses and replaced by a
new participant because of very long overall reading times (more
than twice the group average).
Stimuli
. The stimuli were adapted from those used by Swaab,
Camblin, and Gordon (2004), with two important changes: (1) The
pronouns were replaced by “new names,and (2) a locative phrase
was placed at the start of every sentence so that the antecedent
names would not occur in sentence-initial position (where posi-
tion effects are strongest). Examples of the stimuli are presented
in Example 2 above and in the Appendix; a full set of the materi
-
als is available from the authors. For the companion ERP experi-
ment (Experiment 2), 160 experimental sentences were generated;
of these, 40 were selected for the eye-tracking experiment. Four
versions of each of the experimental sentences were constructed
by manipulating two factors: prominence of NP1 and critical name
type. The sentential subject consisted of either a single proper name
(prominent condition) or two proper names conjoined by and (non-
prominent condition). The second clause began with a temporal or
a causal connective (e.g., when, after), followed by a repetition of
the first-mentioned character in NP1 or a new name. The average
length of the sentences was 14.56 words in the single-NP condition
(range across all conditions, 10–22 words). The critical names were
selected from a list of names that had been generated for previous
studies (Gordon et al., 1999; Swaab et al., 2004). The length of the
critical names varied between five and nine characters. Each name
occurred only once within a stimulus set. Line breaks were placed in
sentences in such a way that the critical name used as the subject of
the second clause never appeared within two words of the beginning
or the end of a line.
We conducted two offline pretests of our experimental materials;
in anticipation of the ERP experiment (Experiment 2), both were
conducted using the entire set of 160 experimental items. The par-
ticipants were native English speakers who did not participate in the
eye-tracking and ERP experiments.
Offline test of stimulus plausibility. To ensure that the replacement
of the repeated names with new names resulted in acceptable sen-
tences, the experimental items were pretested for plausibility with
40 participants. The participants were given one of four counterbal-
anced stimulus sets (each of which contained 40 sentences of each
of the four experimental types) and were asked to rate the sentences
on a scale of 1 (does not make sense) to 5 (makes perfect sense) (see
Table 1).
An ANOVA revealed a significant interaction between the two
experimental factors [F(1,39) 5 19.61, p , .001]. Paired compari-
sons indicated that the ratings for the sentences in the single-NP1/
repeated-name condition were significantly lower than ratings for the
sentences in the conjoined-NP1/repeated-name condition [t(39) 5
3.66, p , .001], the single-NP1/new-name condition [t(39) 5 4.38,
p , .001], and the conjoined-NP1/new-name condition [t(39) 5
3.05, p 5 .004]. This is not surprising, given that prior judgment
(Gordon & Hendrick, 1997; Hudson, Tanenhaus, & Dell, 1986),
behavioral (Almor, 1999; Garrod et al., 1994; Gordon et al., 1993;
Gordon et al., 1999; Kennison & Gordon, 1997; Yang et al., 2003),
and electrophysiological (Swaab et al., 2004) work with repeated-
name coreference has shown that repeated names with prominent
antecedents are difficult to process. Critically, ratings for the sen-
tences in the remaining conditions did not differ from each other,
suggesting that repeated and new names were equally plausible in
the discourse context.
Offline test of the interpretation of repeated names. The results
of the plausibility pretest showed lower ratings for the single-NP1/
repeated-name condition, a finding that is not surprising given pre-
vious evidence that such a configuration of referential NPs is not fe-
licitous. Our second offline pretest used paraphrasing to test whether
repeated NPs in the stimulus sentences are ultimately interpreted
as coreferential (Table 2). The participants were given a sheet with
experimental sentences, with a space after each sentence in which
they were instructed to write a paraphrase of the sentence. The 160
experimental stimuli were divided into eight lists, each containing
20 experimental sentences. Each list was presented in four counter-
balanced versions with 5 stimuli of each type, creating 32 unique
questionnaires, which were completed by 32 participants. The writ-
ten paraphrases for both repeated-name conditions were coded ac-
cording to the attribution of the actions described in the two clauses,
indicating whether the repeated name was interpreted coreferen-
tially or as the introduction of a new character into the discourse. For
some responses, interpretation of the repeated name was ambigu-
ous, either because the two actions were not attributed to a specific
Table 1
Participant Ratings of Stimuli by Experimental Condition
Mean Pretest Rating
Experimental Condition (Scale of 1–5)
Single NP1/repeated name 3.89
Conjoined NP1/repeated name 4.28
Single NP1/new name 4.25
Conjoined NP1/new name 4.22
Note—NP1, first noun phrase.
Table 2
Number and Percentage of Types of Coded Responses in the Paraphrasing Test
Ambiguous
Repeated Both Actions
Same Person
New Person Name Not Included Pronoun Ellipses
Condition No. % No. % No. % No. % No. %
Single NP 0 0 6 4 9 6 84 53 61 38
Conjoined NP 1 1 49 31 13 8 92 58 6 4
Note—NP, noun phrase.
806 Ledoux, Gordon, CambLin, and Swaab
person (e.g., The car was coming when Andrea crossed the street)
or because a repeated name was used without an indication that the
repeated name designated a new person, so that a pronoun could be
substituted for the repeated name without introducing a semantic or
grammatical anomaly (e.g., Andrea saw the car right after Andrea
started to cross the street). Importantly, the overwhelming use of
pronouns and ellipses in the single-NP condition indicated that the
participants were interpreting these repeated names as coreferential,
even though plausibility pretests had shown that this phrasing was
deemed less plausible by another group of participants.
Design and Procedure
. We implemented a 2 3 2 factorial de-
sign, with the factors of NP1 prominence (singular vs. conjoined)
and critical name type (repeated vs. new) manipulated within sub-
jects. The stimuli were counterbalanced across conditions so that no
participant saw a stimulus sentence in more than one condition but,
across participants, every sentence occurred in all four conditions
with equal frequency. There was an initial warm-up block consisting
of 16 filler trials. This was followed by five experimental blocks, each
of which had eight experimental sentences (two stimuli from each of
the four conditions) randomly intermixed with 14 filler trials.
The stimuli were presented on a computer screen, with eye move-
ments measured using an EyeLink system (SensoMotoric Instru-
ments), a head-mounted eyetracker that samples pupil location at a
rate of 250 Hz and parses the samples into fixations and saccades. At
the start of the session, the eyetracker was calibrated for each partici-
pant. Trials began with a fixation point at the location where the first
word of the sentence would subsequently be presented. The fixation
point served to focus the participant’s attention at the correct loca-
tion and also to allow the experimenter, using a second computer, to
monitor the location and steadiness of the participant’s gaze. The ex-
perimenter initiated the trial when the participant was appropriately
fixated, causing the stimulus sentence to appear. The participant
was instructed to read the sentence in a natural way, not to hurry, but
not to linger excessively. He or she was asked to press the space bar
when finished. Following each sentence, a true/false comprehension
question appeared; the participant responded by pressing labeled
keys. The experimenter’s computer provided an online display of the
participant’s fixations. When these displays started to exhibit drift,
the experimenter paused the experiment to recalibrate the tracker.
Results
Eye tracks were analyzed to provide the following stan-
dard measures of reading time (see Liversedge, Paterson,
& Pickering, 1998; Rayner, 1978, 1998): (1) first-fixation
duration; (2) gaze duration (the sum of the durations of the
fixations on a region of interest from the first time that re-
gion is fixated until a region outside the region of interest
is fixated, provided that the eyes have not yet gone beyond
the region of interest); (3) total time (the sum of all fixa-
tion durations on a region of interest); and (4) rereading
duration (the difference between total reading time and
gaze duration). Table 3 shows these measures for the fol-
lowing regions of interest: (1) the first name that appeared
in the sentence, which is a possible target of rereading
when trying to understand possible coreference relations;
(2) the word preceding the critical name in the second
clause, which provides evidence about any overall differ-
ence in difficulty due to the type of subject NP for the first
clause and about possible effects of parafoveal preview of
the following critical name; (3) the critical name in the
second clause, which provides evidence about repetition
priming and, possibly, about coreferential interpretation;
and (4) the verb of the second clause (including auxiliary
verbs, quantifiers, gerunds, and infinitival complements),
which provides information about nonimmediate, possi-
bly integrative effects of coreferential interpretation.
First name
. There were no significant main effects or
interactions on any of the reading time measures for the
first name in the sentence.
Word before critical name
. For first-fixation dura-
tions, times were significantly shorter in the conjoined
than in the singular condition for participants, but not for
items [F
1
(1,39) 5 5.29, p , .05; F
2
(1,39) , 1]. Neither
the effect of type of critical name nor the interaction of
type of name with NP1 prominence approached signifi-
cance. For gaze duration, times were again significantly
shorter in the conjoined than in the singular condition for
participants, but not for items [F
1
(1,39) 5 4.74, p , .05;
F
2
(1,39) 5 3.58, p , .07]. Again, neither the effect of
type of critical name nor the interaction of type of name
with NP1 prominence approached significance. Total
reading times for this region were marginally shorter in
the conjoined condition than in the singular condition in
the participants analysis [F
1
(1,39) 5 3.87, p , .06] and
significantly shorter in the items analysis [F
2
(1,39) 5 7.3,
Table 3
Eye-Tracking Results of Experiment 1 (in Milliseconds)
Measure
Region of Interest Condition First Fixation Gaze Total Rereading
First name Singular/repeated 197 232 437 223
Conjoined/repeated 196 234 425 203
Singular/new 193 226 411 204
Conjoined/new 196 236 411 196
Critical name minus one Singular/repeated 195 227 374 186
Conjoined/repeated 190 216 326 157
Singular/new 197 229 345 172
Conjoined/new 185 210 338 172
Critical name Singular/repeated 181 190 308 134
Conjoined/repeated 182 193 274 107
Singular/new 196 215 332 143
Conjoined/new 191 214 323 137
Verb region Singular/repeated 279 466 199
Conjoined/repeated 263 407 158
Singular/new 273 421 165
Conjoined/new 283 453 184
CoreferenCe and LexiCaL repetition 807
p , .01]. There was no main effect of type of critical name
(repeated vs. new), nor was there a significant interaction
of type of NP1 and type of critical name. For rereading, no
effects approached statistical significance.
In sum, reading time measures for the word before the
critical word showed some evidence of facilitated pro-
cessing in the conjoined condition relative to the singular
condition, although this effect was not significant in all
the analyses. One possible explanation of the trend is that
reading speed increases as a reader progresses further into
a text (Ferreira & Henderson, 1995) and this word is fur-
ther from the beginning of the sentence in the conjoined
than in the singular condition.
Critical name
. For first-fixation duration, times
were shorter for repeated names than for new names
[F
1
(1,39) 5 10.56, p , .005; F
2
(1,39) 5 10.86, p ,
.005], with other effects not approaching significance.
Likewise for gaze duration, times were shorter for re-
peated names than for new names [F
1
(1,39) 5 22.56, p ,
.001; F
2
(1,39) 5 19.09, p , .001], with other effects not
approaching significance. Total reading times on the criti-
cal name were shorter for repeated names than for new
names [F
1
(1,39) 5 17.91, p , .001; F
2
(1,39) 5 10.07,
p , .005]. They were also shorter for the conjoined con-
dition than for the singular condition [F
1
(1,39) 5 7.30,
p , .01; F
2
(1,39) 5 5.08, p , .05]. The interaction of
these factors was not significant. For rereading, there was
a trend toward shorter times for repeated names than for
new names [F
1
(1,39) 5 4.28, p , .05; F
2
(1,39) 5 3.38,
p , .08] and for sentences with conjoined subjects than
for those with singular subjects [F
1
(1,39) 5 3.77, p , .06;
F
2
(1,39) 5 3.86, p , .06]. The interaction of these factors
was not significant [F
1
(1,39) 5 1.65, p , .21; F
2
(1,39) 5
1.56, p , .25].
In sum, reading times for the critical name show highly
reliable evidence of repetition priming (shorter times for
repeated names than for new names) in measures that re-
flect early lexical processing (first-fixation duration and
gaze duration). This difference persists, to some extent,
in later measures (i.e., rereading) and comprehensive
measures (i.e., total reading time) of processing. Read-
ing times for the critical name also showed shorter total
durations for sentences with plural NP1 subjects than for
those with singular NP1 subjects, a pattern that is similar
to what was found for the preceding word.
Verb region
. First-fixation duration was not analyzed
for this region because, unlike the other regions, it could
consist of more than one word, which greatly limits the
meaning of first-fixation duration as a measure of pro-
cessing for the entire region. No significant effects were
observed for this region for gaze duration measures. Total
reading times for the verb region did not show signifi-
cant main effects of either the type of critical name or the
type of NP1 prominence. However, there was a significant
crossover interaction in total reading times between these
factors: For repeated names, times were shorter when the
subject of the first clause was a conjoined NP than when
it was a singular NP; for new names, times in these condi-
tions showed the reverse pattern [F
1
(1,39) 5 5.18, p ,
.05; F
2
(1,39) 5 7.22, p , .05]. For rereading, there were
no significant main effects, but the interaction between
type of critical name and NP1 prominence was significant
[F
1
(1,39) 5 5.53, p , .05; F
2
(1,39) 5 4.78, p , .05].
In sum, the verb region shows evidence in a compre-
hensive processing measure (i.e., total reading time) and a
measure of later processing (i.e., rereading) that repeated-
name coreference to a prominent entity (a singular sub-
ject) creates difficulty in comprehension.
Discussion
In this first experiment, in which the participants read
normally while their eye movements were monitored, we
found a dissociation between effects of lexical repetition
and effects of discourse context. We found evidence of
repetition priming on early measures of reading at the crit-
ical word. First-fixation duration and gaze duration mea-
sures of the critical word were shorter to repeated names
than to new names. In line with results from text repetition
studies, the benefit conferred by repetition persisted, to
some extent, in a later measure (rereading) and in a com-
prehensive measure (total reading time). The early emer-
gence of this effect suggests a facilitation by repetition of
processes of lexical access. The persistence of this effect
suggests that repetition might also ease some aspects of
discourse integration.
Critically, the discourse manipulation of the prominence
of the first NP influenced the ease of establishing corefer-
ence with repeated names, as was demonstrated by the
interaction of type of first NP and type of name on depen-
dent measures reflecting later processing of the region fol-
lowing the critical name. When a repeated name followed
a prominent antecedent, total reading times and rereading
times on the verb region following the critical name were
longer than those in the nonprominent condition. That
this effect emerged in later durational measures suggests
that it stems from integrative processes in language com-
prehension, not from processes of lexical access. These
results provide another example of the repeated-name
penalty described in previous behavioral work (Almor,
1999; Garrod et al., 1994; Gordon et al., 1993; Gordon
et al., 1999; Kennison & Gordon, 1997; Yang et al., 2003)
and are, thus, consistent with the mechanisms described
by the discourse prominence theory (Gordon & Hendrick,
1998b).
EXPERIMENT 2
In the second experiment, we measured ERPs to the
critical name in sentences such as those in Example 2 in
order to examine the same factors of discourse promi-
nence and repetition as in Experiment 1. Here, we will
begin by reviewing the ERP effects of lexical repetition in
word lists. We will then describe the handful of studies in
which lexical repetition in sentence or discourse contexts
has been examined.
Lexical Repetition and ERPs
The effects of repetition on ERPs to words in lists
are well established, being most strongly linked to two
ERP components, the N400 and the late positive com-
808 Ledoux, Gordon, CambLin, and Swaab
plex (LPC). The N400 is a negative deflection in the
ERP waveform that peaks approximately 400 msec
poststimulus-onset and is maximal over posterior elec-
trode sites. A reduction of the amplitude of the N400 is
found to words that can be easily integrated into the pre-
ceding word, sentence, or discourse context; this compo-
nent is thus sensitive to processes of lexical integration
(e.g., Brown & Hagoort, 1993; Chwilla, Brown, & Ha-
goort, 1995; Holcomb, 1993; Rugg, Furda, & Lorist, 1988;
van Berkum, Hagoort, & Brown, 1999; Van Petten &
Kutas, 1991). The LPC is a positive-deflecting component
that begins approximately 400 msec post-stimulus-onset
and can continue until 900 msec. It is also maximal over
posterior electrode sites and is sensitive to explicit recall
(Paller & Kutas, 1992; Paller, Kutas, & McIsaac, 1995;
Rugg, 1985, 1990; Swick & Knight, 1997).
Repetition of words in lists leads to a reduction in the
amplitude of the N400 and an increase in the amplitude of
the LPC. A reduction of the N400 for repeated words in
lists is said to reflect the processing advantage provided by
repetition. This N400 reduction has been shown to persist
across lags; however, the effect is greatest at the shortest
lags (Nagy & Rugg, 1989). The increase of the LPC for
repeated words in lists (relative to nonrepeated controls)
has been linked to the explicit recall of the prior presenta-
tion (Paller et al., 1995).
Lexical Repetition in Discourse
A handful of studies have examined the electrophysi-
ological response to words that are repeated within a
sentence or discourse context. Even when this repetition
occurs for reasons other than that of establishing corefer-
ence, the electrophysiological signature of the effects of
lexical repetition in discourse contexts can differ from that
for word lists. Van Petten, Kutas, Kluender, Mitchiner, and
McIsaac (1991) studied lexical repetition effects in dis-
course, using passages taken from the Reader’s Digest. As
in word lists, repetitions of both content words and proper
names resulted in N400 amplitude reductions. However,
the LPC to these two types of words was differentially
affected by repetition; whereas the LPC to proper names
was more positive with repetition, it was reduced for re-
peated content words. Van Petten et al., interpreted these
modulations of the LPC as reflecting the differential re-
trieval and updating demands made by content words and
proper names in a discourse context. To the extent that se-
mantic representations activated during the initial presen-
tation of content words were still active at the time of the
repetition, semantic retrieval processes were facilitated,
resulting in a decrease in LPC amplitude for these words
upon repetition. Because the proper names in this study
referred to people who were not likely to be known to the
participants, their initial presentation required little in the
way of retrieval processes, since there was no prior infor-
mation available to be retrieved. On subsequent presenta-
tions of the proper name, comprehension depended on the
reinstatement of the prior memory representation and its
updating by new discourse information. Thus, the ampli-
tude of the LPC, as a marker of retrieval and updating,
is reduced for repetitions of content words in discourse
(because these words require less such processing on sub-
sequent presentations) but increased for repeated proper
names (because these words require more such processing
on subsequent presentations). It is important to note that
in Van Petten et al.s study, most instances of coreference
were established by the use of pronouns. It was only after
a referent had been out of discourse focus for some time
that a name was repeated, suggesting that these instances
are better thought of as cases of reinstatement (O’Brien,
Albrecht, Hakala, & Rizzella, 1995), rather than of local
coreference.
One recent study did examine lexical priming effects
and discourse coreference effects during sentence process-
ing. Anderson and Holcomb (2005) looked at the ERPs
to NPs in two-sentence contexts; an example is shown in
Example 5:
5. First sentence: Kathy sat nervously in the cab/
taxi on her way to the airport.
Second sentence: The cab came very close to hit-
ting the car.
A cab came very close to hit-
ting the car.
Coreference was manipulated by the use in the second
sentence of a definite NP (beginning with the word the)
or an indefinite NP (beginning with the word a), used to
refer back to a noun that had been introduced as an object
in the first sentence. Priming was manipulated by the use
in the second sentence of a repetition of the critical noun
from the first sentence or a synonymous word. Anderson
and Holcomb (2005) reported N400 priming effects at the
critical word for both repetitions and synonyms; consis-
tent with previous studies, repetition-priming effects were
greater than semantic-priming effects, but neither had an
effect on the LPC. However, there was no evidence of a
main effect of coreference on the N400, nor was there an
interaction of the two factors on this component. The ab-
sence of an interaction in this case between repetition and
coreference is consistent with work related to the Gordon
and Hendrick (1998b) model, because the antecedent ex-
pressions in this study were not syntactically prominent in
the discourse (e.g., in a postverbal adjunct phrase).
Swaab et al. (2004) used ERPs to examine the establish-
ment of coreference with pronouns and repeated names.
They found an N400 reduction to repeated names that were
preceded by a nonprominent referent, relative to those pre-
ceded by a prominent referent. Because the N400 is sen-
sitive to difficulties in lexical integration, this difference
suggests that the difficulty of achieving coreference with a
name increased with the prominence of the referent.
However, in Swaab et al.s (2004) experiment, along
with those of Gordon et al. (1999), pronouns were used as a
comparison for repeated names because this allowed a test
of theoretical notions about the centrality of pronominal
reference in discourse coherence. It is important to note,
however, that names and pronouns have been shown to dif-
fer along several dimensions, such as frequency (with pro-
nouns being more frequent in the language), length (with
pronouns tending to be shorter than proper names), and
CoreferenCe and LexiCaL repetition 809
word class (pronouns are considered closed-class words).
All of these factors have been shown to affect language
processing, both behaviorally and electrophysiologically
(for reviews, see Hauk & Pulvermüller, 2004; Osterhout,
Allen, & McLaughlin, 2002). For this reason Swaab et al.s
(2004) experiment provided no evidence with regard to
the interaction of lexical repetition effects with discourse
prominence. In the present experiment, noncoreferential
new names were used to directly measure lexical repetition
effects and coreferential processes during reading.
Using the conditions shown in Example 2, Experi-
ment 2 examined the interaction of lexical repetition and
discourse prominence in order to test the prediction that
coreference to a prominent antecedent causes a repeated
name to be processed as if it were a new name. Previous
ERP research has demonstrated a reduction in the ampli-
tude of the N400 to repeated words, in word lists and in
sentence contexts; we therefore might expect to find a re-
duced N400 to the critical repeated name (relative to a new
name) in sentences that are otherwise identical (i.e., in 2a
vs. 2b and in 2c vs. 2d). We predicted, however, that this
repetition effect would be modulated by sentence context.
Specifically, we expected that the difficulty of establishing
coreference between a repeated name and a prominent an-
tecedent would be reflected as an increase in the difficulty
of integration processes in the singular-NP1/repeated-
name condition (2a), relative to the conjoined-NP1/
repeated-name condition (2c), in which both repetition
and ease of establishing coreference would work to reduce
the amplitude of the N400. The effects of repetition on
the LPC have varied in sentential contexts; difference in
this experiment might be informative about differential
retrieval and updating demands in these sentences.
Method
Participants
. The participants were 20 right-handed native
speakers of English. All had normal or corrected-to-normal vision
and were recruited from the university population of the University
of California at Davis. None of the participants reported neurologi-
cal impairment, and none were currently taking medication. The
participants gave informed consent before the experiment and were
compensated with payment or with course credit.
Materials
. The stimuli consisted of all of the 160 experimental
items that had been pretested as described above (see Experiment 1).
Each of the stimuli appeared in the same four conditions as in the
previous experiment, defined by the crossing of type of subject of
the initial clause (singular or conjoined) and type of subject of the
second clause (repeated or new name).
Ninety filler sentences were created that also contained named
characters but that had structures different from those of the experi-
mental sentences in order to mitigate participant strategies.
Design and Procedure
. Ten filler sentences formed an initial
practice block, which served to familiarize the participants with the
stimulus presentation and task. The 160 experimental sentences were
pseudorandomly mixed with the remaining 80 filler sentences into
eight subsequent blocks of 30 sentences each. The first three items
and the final item in each experimental block were filler sentences.
Four groupings of the experimental sentences were constructed so
that a given participant read each experimental sentence once and
read equal numbers of sentences in each of the conditions. Across
participants, each passage occurred equally often in each condition.
Each participant was tested individually in a dimly lit, electrically
shielded, sound-attenuating booth. They were seated in a comfort-
able chair approximately 100 cm from the computer screen. They
were asked to silently read the sentences and to answer a true/false
comprehension question after each sentence by pressing one of two
buttons on a button box.
Each trial began with a fixation cross that was presented for
1,000 msec in the center of the screen, to alert the participants to the
beginning of the trial. The fixation cross was replaced by the first
word of the sentence, which was replaced by subsequent words in
typical rapid serial visual presentation fashion. Each sentence was
presented for a duration of 300 msec per word, with an interstimulus
interval of 200 msec. Characters appeared as white letters against a
dark background in 14-point Tahoma font. The first word of each
sentence and all the proper names began with a capital letter; the last
word was presented together with a period. During the presentation
of the experimental and filler items, the participants were asked to
refrain from moving their eyes and from blinking. The true/false
comprehension question appeared all at once 1,000 msec after the
last word of the sentence and remained on the screen until a response
was recorded. Once the participants had responded to the true/false
statement, a prompt appeared on the screen, and the participants
started the next trial by pressing a button. The participants were al-
lowed a short break after each block.
EEG recording
. EEG was recorded from 29 tin electrodes fit-
ted in an elastic cap (see Figure 1), referenced to the right mastoid.
Vertical eye movements were monitored by a suborbital electrode,
and horizontal eye movements via left and right external canthus
montages. Impedance was kept below 5 k
W. Prior to offline aver-
aging, all single-trial waveforms were automatically screened for
amplifier blocking, muscle artifacts, horizontal eye movements, and
blinks over epochs of 1,200 msec, starting 200 msec before the onset
of the critical words. For each participant, average ERPs were com-
puted over artifact-free trials for critical words in all four conditions.
Off line, the waveforms were rereferenced to the algebraic average
of both mastoids. The bandpass was 0.01–30 Hz at a sampling rate
of 250 Hz.
Results
ERP data were analyzed using repeated measures ANO-
VAs performed on the mean amplitude of the ERPs to
the critical words over the 29 electrode sites in the N400
(250- to 500-msec) and LPC (500- to 700-msec) time
windows (relative to a 100-msec prestimulus baseline).
In each time window, an omnibus analysis was first con-
ducted over three independent variables, all of which
were tested within subjects: NP1 prominence (with two
levels: singular vs. conjoined), critical name type (with
two levels: repeated vs. new), and electrode site (with 29
levels). Significant interactions were tested with subse-
quent contrasts, as described below. For evaluating effects
with more than one degree of freedom in the numerator,
the Greenhouse–Geisser correction was used to compen-
sate for inhomogeneous variances and covariances across
treatment levels (Greenhouse & Geisser, 1959); the ad-
justed p values are reported.
Figures 2 and 3 show the grand average ERPs in the four
conditions. In Figure 2, the plots were generated to high-
light the effects of the discourse manipulation, showing the
grand average ERPs to the critical repeated (left panel) and
new (right panel) names in the two prominence conditions.
In Figure 3, the plots were generated to highlight the effects
of the repetition manipulation, showing the grand average
ERPs to the critical names in the prominent condition (left
panel) and in the nonprominent condition (right panel).
N400 time window
. An omnibus ANOVA performed
on the 250- to 500-msec epoch revealed a significant in-
810 Ledoux, Gordon, CambLin, and Swaab
teraction between NP1 prominence and critical name type
[F(1,19) 5 5.03, p 5 .037]. Paired comparisons demon-
strated a significant difference for the N400 by promi-
nence condition for the repeated names [F(1,19) 5 4.79,
p 5 .041], but not for the new names (F , 1). A significant
difference for the N400 by name condition was found for
the nonprominent condition [F(1,19) 5 7.95, p 5 .011],
but not for the prominent condition (F , 1).
LPC time window
. As in Swaab et al.s (2004) analy-
sis, the interaction between discourse focus and critical
name type did not reach significance in this time window
[F(1,19) 5 1.15, p 5 .338]. However, because Swaab et al.
found a significant LPC effect over posterior electrodes for
repeated names, we conducted planned comparisons over
these sites as well. For the repeated names, there was no
significant effect of discourse focus (F , 1), but there was
a significant interaction between this factor and electrode
site [F(28,532) 5 2.43, p 5 .04], demonstrating a greater
positivity for repeated names following a nonprominent
antecedent over posterior electrodes. There was no differ-
ence for the new names in this time window (Fs , 1).
2
Discussion
In this experiment, repeating a word as part of a coref-
erential relationship had an ERP effect similar to that seen
in other situations involving repeated words—as long as
the antecedent of the repeated word did not have linguistic
prominence. In the nonprominent condition, a repeated
coreferential name elicited a reduced N400, relative to a
lexically matched new name. The LPC to these words was
not influenced by repetition, a finding that differs from
that of Van Petten et al. (1991). It is possible that the read-
ing of the repeated and new names in this experiment led
equally to the engagement of retrieval and updating pro-
cesses. It seems, then, that processes of coreference (as
studied in this experiment) and processes of reinstatement
(as examined by Van Petten et al., 1991) may differentially
affect the amplitude of the LPC.
Importantly, when a repeated name was coreferential
with a linguistically prominent antecedent, the standard
N400 repetition effect was not observed. When the subject
of the first clause of the sentence consisted of a single
name, the N400 elicited by a repeated coreferential name
did not differ from that elicited by a new name. It is not
the case that words in a discourse invariably benefit from
repetition; this benefit may be modulated by factors that
are unique to the structure of the discourse.
Repeated coreferential names that followed a promi-
nent antecedent elicited a larger N400 than did identical
repeated coreferential names that followed a nonpromi-
nent antecedent, an electrophysiological manifestation
of the repeated name penalty that has been described in
previous behavioral work (Almor, 1999; Garrod et al.,
1994; Gordon et al., 1993; Gordon et al., 1999; Kennison
& Gordon, 1997; Yang et al., 2003). The N400 effect of
discourse prominence was accompanied here (as in Swaab
et al., 2004) by a difference to the repeated names in the
LPC epoch. Repeated names that followed a nonpromi-
Figure 1. Configuration of the electrodes in the cap.
RM
(Ref)
LM
FP1
AFz
FP2
Fz
FC1
FC2
F7
F3
FC5
T3
C3
CP5
T5
P3
Cz
CP1
CP2
Pz
POz
O1
O2
F8
gnd
F4
FC6
C4
T4
CP6
P4
T6
CoreferenCe and LexiCaL repetition 811
nent antecedent elicited a greater positivity in this window
than did the same repeated names when they followed a
prominent antecedent. If the LPC reflects retrieval and
updating processes, it seems that repeated names initiate
these processes to a greater degree following a conjoined
(relative to a singular) sentential subject. It seems likely
that the conjoined sentential subject initially imposed a
larger working memory load; the LPC difference here,
then, may reflect the greater demands of reinstating and
updating a larger memory load (Van Petten et al., 1991).
These findings are consistent with the mechanisms
described in the Gordon and Hendrick (1998b) model.
According to the discourse prominence model, repeated-
name coreference to a prominent entity leads to a period of
disjoint reference, wherein the discourse model contains
two entities that happen to have the same name. The offline
study of paraphrasing indicates that in the vast majority of
cases, participants do ultimately achieve a coreferential
interpretation of the repeated name in the prominent an-
tecedent condition. The present results indicate that this
Figure 2. The effect of discourse prominence on repeated (left panel) and new (right
panel) names. The ERPs are grand averages across all participants, recorded from
frontal (F3, Fz, F4), central (C3, Cz, C4), and posterior (P3, Pz, P4) sites. ERPs were
time-locked to the critical name in the second clause (underlined). The arrow points to
the N400 region on electrode Cz.
Prominent NP1
(At the office, Daniel moved the cabinet . . . )
Nonprominent NP1
(At the office, Daniel and Amanda moved the cabinet . . . )
Repeated Names
New Names
Discourse Effects
( . . . because Robert needed room for the desk.)
F3
FZ
F4
C3
CZ
C4
P3
PZ
P4
F3
C3
P3
FZ
CZ
PZ
F4
C4
P4
0 400 800
2.0 µV
( . . . because Daniel needed room for the desk.)
Figure 3. Repetition effect following a prominent (left panel) and a nonprominent
(right panel) first noun phrase. The ERPs are grand averages across all participants,
recorded from frontal (F3, Fz, F4), central (C3, Cz, C4), and posterior (P3, Pz, P4)
sites. ERPs were time-locked to the critical name in the second clause (underlined).
The arrow points to the N400 region on electrode Cz.
Repeated name
( . . . because Daniel needed room for the desk.)
New name
( . . . because Robert needed room for the desk.)
Prominent NP1
Nonprominent NP1
Repetition Effects
(At the office, Daniel moved the cabinet . . . ) (
At the office, Daniel and Amanda moved the cabinet . . . )
F3
FZ
F4
C3
CZ
C4
P3
PZ
P4
F3
C3
P3
FZ
CZ
PZ
F4
C4
P4
0 400 800
2.0 µV
812 Ledoux, Gordon, CambLin, and Swaab
process is delayed past the time period measured by ERPs
to the name itself.
GENERAL DISCUSSION
When presented with a task involving lists of words,
participants engage in many processes that are similar
to those that are engaged when words in sentences are
read. Processing the words in a sentence or discourse re-
quires the additional step of integrating lexical-semantic
information into the computation of the overall meaning
or context of the sentence or discourse. Coreferential in-
terpretation of NPs is one area of discourse processing
where such integration is critical. The establishment of
coreference is the basis by which the characteristics and
actions associated with different expressions that refer to
the same entity are integrated together (Garnham, 2001;
Gordon & Hendrick, 1998b; Kintsch & van Dijk, 1978;
Sanford & Garrod, 1981).
In two experiments, using eye tracking and ERPs, we
found evidence that the sentential context in which words
were repeated attenuated the magnitude of repetition-
priming effects. Effects of repetition priming were demon-
strated in both experiments: In Experiment 1, durational
measures of early processing (first-fixation duration and
gaze duration) were shorter for repeated words than for
new words, and in Experiment 2, we found a reduction
of the N400 to repeated words, relative to new words.
However, in both experiments, at least at some point in
processing, the magnitude of these repetition-priming
effects varied as a function of the type of NP that was
introduced in the first clause of the sentence. In the eye-
tracking experiment, later processing measures (total
reading time and rereading) for the region following the
critical name showed repetition priming when the first NP
introduced two conjoined entities to the discourse model,
but not when the first NP introduced a single entity. In the
ERP experiment, when the first NP introduced a single
entity, repetition-priming effects were eliminated.
Both experiments also provided evidence that the pro-
cessing of a coreferential repeated name was more diffi-
cult when the antecedent was prominent in the discourse
representation. Readers had more difficulty integrating
the two instances of a repeated name (as indexed by later
processing measures in the eye-tracking experiment and
by the N400 in the ERP experiment) in sentences in which
the first instance was introduced as the singular subject of
the first NP, relative to cases in which the first instance
was embedded in a conjunctive first NP. This difficulty of
integrating a coreferential repeated name with a promi-
nent antecedent is an example of the repeated name pen-
alty that has been described in previous behavioral work
(Almor, 1999; Garrod et al., 1994; Gordon et al., 1993;
Gordon et al., 1999; Kennison & Gordon, 1997; Yang
et al., 2003). We add to this body of research through the
use of ERP and eye-tracking methodologies, which al-
lowed us to more finely dissociate effects of repetition
and prominence than has generally been possible using
the more global behavioral measures of those studies.
More important, the inclusion in the present experiments
of a noncoreferential, lexically matched control for the
repeated names provided a more explicit baseline for mea-
suring repetition effects and coreferential processing than
had been used in previous studies (Swaab et al., 2004),
which have tended to use coreferential pronouns as a se-
mantic control but which have not had adequate controls
for lexical characteristics of the critical words.
These findings support a model of coreference such
as the discourse prominence model (Gordon & Hendrick
1998b), in which factors that are unique to the construc-
tion of a discourse model interact with general mecha-
nisms of memory in determining the mental representa-
tion of a discourse. They show that the ease or difficulty
of establishing coreference using repeated names varies
depending on factors that are unique to the construction of
a discourse representation. In some cases, such as those in
which the antecedent is prominent in the discourse, coref-
erence with repeated names may be exceedingly difficult;
however, our offline paraphrasing study showed that read-
ers do eventually come to take the two instances of the
name to refer to the same entity. The Gordon and Hendrick
(1998b) model proposes that coreferential repeated names
are initially interpreted as introducing new entities to the
discourse model, a suggestion that finds support in the
two experiments described here. The model also proposes
that at some point, further integrative processing leads to
the establishment of the equivalence of the two names and
to an ultimately coreferential interpretation. Whether the
differences seen in the later processing measures or the
amplitude of the N400 to repeated words that followed a
singular antecedent reflect the positing of a new entity in
the discourse model only or the additional process of es-
tablishing equivalence between two entities predicated on
the same name is impossible to determine at this point.
A number of studies have compared the results from
ERPs with those from eye tracking (Camblin, Gordon,
& Swaab, 2007; Deutsch & Bentin, 2001; Gordon et al.,
2004; Sereno, Rayner, & Posner, 1998), an approach that
has been advocated in the literature (Kutas & Federmeier,
1998; Sereno & Rayner, 2003). Our use of a subset of the
ERP stimuli in the eye-tracking experiment allows such a
comparison. Perhaps the most notable difference between
the results of the two experiments is the locus of the ma-
jority of the significant effects. In the ERP experiment, we
found evidence of repetition priming and the interaction
of the repetition and discourse factors in the ERPs that
were time-locked to the critical name in the second clause
of the sentences. In the eye-tracking experiment, we found
a different pattern of results; although we found evidence
of repetition priming on the critical word itself, evidence
of the interaction of this factor with discourse prominence
was displaced to the region following the critical word. In
addition, in the eye-tracking experiment, the effect of rep-
etition was seen to influence (primarily) early processing
measures, whereas the discourse effect was seen to influ-
ence later processing measures; in the ERP experiment,
both of these factors influenced the same component (and
thus, a similar time course of expression). These differ-
ences may result from basic methodological differences
between the ERP and the eye-tracking paradigms. In our
CoreferenCe and LexiCaL repetition 813
ERP experiment, as in many such studies in which reading
has been examined, we used rapid serial visual presenta-
tion with a relatively slow presentation rate of 500 msec
per word to allow better resolution of the ERP compo-
nents of interest. However, this means that the reader is
made to look at each word longer than the typical fixation
rate during natural reading of approximately 250 msec.
This is not the case in the eye-tracking experiment, in
which readers have control of where they are looking and
for how long. The prolonged presentation duration in the
ERP paradigm may give the reader enough time to engage
in processes (such as those of coreferential integration)
that would normally be extended over a region of several
words in natural reading. Indeed, recent models of eye
movements during reading describe processing as distrib-
uted in nature, in that a given fixation on a word might
reflect the simultaneous contributions from the processing
of the previous, current, and subsequent words (Kliegl,
Nuthmann, & Engbert, 2006). In the eye-tracking results,
we may be seeing processes of lexical access (priming)
being localized to the critical word itself, but processes of
sentential integration (as influenced by discourse promi-
nence) spread over a number of words.
In summary, we measured eye movements and ERPs to
coreferential repeated names (and noncoreferential new
name controls) in order to examine mechanisms of dis-
course integration. Processing the coreferential relation-
ship between two expressions requires the establishment
and maintenance of a representation of the information con-
veyed by the antecedent expression, followed by retrieval of
that information on the basis of cues in the coreferential ex-
pression. These processes of representation and retrieval are
strongly influenced by syntactic and discourse structure.
AUTHOR NOTE
This research was supported by NIMH Grant RO1 MH066271 to
P.C.G. and T.Y.S. We thank Scott Hajek and Stephanie Moser for assis-
tance with data collection. Correspondence concerning this article should
be addressed to K. Ledoux, Cognitive Neurology/Neuropsychology,
Johns Hopkins University, 1629 Thames St., Suite 350, Baltimore, MD
21231 (e-mail: kledoux1@jhmi.edu).
REFERENCES
Albrecht, J. E., & Clifton, C., Jr. (1998). Accessing singular anteced-
ents in conjoined phrases. Memory & Cognition, 26, 599-610.
Almor, A. (1999). Noun-phrase anaphora and focus: The informational
load hypothesis. Psychological Review, 106, 748-765.
Anderson, J. E., & Holcomb, P. J. (2005). An electrophysiological
investigation of the effects of coreference on word repetition and syn-
onymy. Brain & Language, 94, 200-216.
Brown, C. M., & Hagoort, P. (1993). The processing nature of the
N400: Evidence from masked priming. Journal of Cognitive Neuro-
science, 5, 34-44.
Camblin, C. C., Gordon, P. C., & Swaab, T. Y. (2007). The interplay
of discourse congruence and lexical association during sentence pro-
cessing: Evidence from ERPs and eye tracking. Journal of Memory &
Language, 56, 103-128.
Carreiras, M. (1997). Plural pronouns and the representation of their
antecedents. European Journal of Cognitive Psychology, 9, 53-87.
Chomsky, N. (1981). Lectures on government and binding. Dordrecht:
Foris.
Chwilla, D. J., Brown, C. M., & Hagoort, P. (1995). The N400 as a
function of the level of processing. Psychophysiology, 32, 274-285.
Deutsch, A., & Bentin, S. (2001). Syntactic and semantic factors in
processing gender agreement in Hebrew: Evidence from ERPs and
eye movements. Journal of Memory & Language, 45, 200-224.
Ferreira, F., & Henderson, J. M. (1995). Reading processes during
syntactic analysis and reanalysis. In J. M. Henderson, M. Singer, &
F. Ferreira (Eds.), Reading and language processing (pp. 119-147).
Mahwah, NJ: Erlbaum.
Garnham, A. (2001). Mental models and the interpretation of anaphora.
Hove, U.K.: Psychology Press.
Garrod, S., Freudenthal, D., & Boyle, E. A. (1994). The role of
different types of anaphor in the on-line resolution of sentences in a
discourse. Journal of Memory & Language, 33, 39-68.
Gordon, P. C., Camblin, C. C., & Swaab, T. Y. (2004). On-line mea-
sures of coreferential processing. In M. Carreiras & C. Clifton (Eds.),
The on-line study of sentence comprehension: Eyetracking, ERPs, and
beyond (pp. 139-150). New York: Psychology Press.
Gordon, P. C., Grosz, B. J., & Gilliom, L. A. (1993). Pronouns,
names, and the centering of attention in discourse. Cognitive Science,
17, 311-347.
Gordon, P. C., & Hendrick, R. (1997). Intuitive knowledge of linguis-
tic coreference. Cognition, 62, 325-370.
Gordon, P. C., & Hendrick, R. (1998a). Dimensions of grammatical
coreference. In M. A. Gernsbacher & S. J. Derry (Eds.), Proceedings
of the Twentieth Annual Conference of the Cognitive Science Society
(pp. 424-429). Hillsdale, NJ: Erlbaum.
Gordon, P. C., & Hendrick, R. (1998b). The representation and pro-
cessing of coreference in discourse. Cognitive Science, 22, 389-424.
Gordon, P. C., Hendrick, R., Ledoux, K., & Yang, C. L. (1999). Pro-
cessing of reference and the structure of language: An analysis of com-
plex noun phrases. Language & Cognitive Processes, 14, 353-379.
Greenhouse, S. W., & Geisser, S. (1959). On methods in the analysis
of profile data. Psychometrika, 24, 95-112.
Grosz, B. J., Joshi, A. K., & Weinstein, S. (1995). Centering: A frame-
work for modeling the local coherence of discourse. Computational
Linguistics, 21, 203-225.
Hauk, O., & Pulvermüller, F. (2004). Effects of word length and
frequency on the human event-related potential. Clinical Neurophysi-
ology, 115, 1090-1103.
Holcomb, P. J. (1993). Semantic priming and stimulus degradation:
Implications for the role of the N400 in language processing. Psycho-
physiology, 30, 47-61.
Hudson, S. B., Tanenhaus, M. K., & Dell, G. S. (1986). The effect of
the discourse center on the local coherence of a discourse. In Proceed-
ings of the 8th Annual Conference of the Cognitive Science Society
(pp. 96-101). Hillsdale, NJ: Erlbaum.
Jacoby, L. L., & Dallas, M. (1981). On the relationship between auto-
biographical memory and perceptual learning. Journal of Experimen-
tal Psychology: General, 110, 306-340.
Johnson-Laird, P. N. (1983). Mental models: Toward a cognitive sci-
ence of language, inference, and consciousness. Cambridge, MA:
Harvard University Press.
Kamp, H., & Reyle, U. (1993). From discourse to logic: Introduction
to model theoretic semantics of natural language, formal logic, and
discourse representation theory. Dordrecht: Kluwer.
Kaup, B., Kelter, S., & Habel, C. (2002). Representing referents of
plural expressions and resolving plural anaphors. Language & Cogni-
tive Processes, 17, 405-450.
Keller, F., & Asudeh, A. (2001). Constraints on linguistic coreference:
Structural vs. pragmatic factors. In J. D. Moore & K. Stenning (Eds.),
Proceedings of the 23rd Annual Conference of the Cognitive Science
Society (pp. 483-488). Mahwah, NJ: Erlbaum.
Kennison, S. M., & Gordon, P. C. (1997). Comprehending referential
expressions during reading: Evidence from eye tracking. Discourse
Processes, 24, 229-252.
Kintsch, W., & van Dijk, T. A. (1978). Toward a model of text compre-
hension and production. Psychological Review, 85, 363-394.
Kliegl, R., Nuthmann, A., & Engbert, R. (2006). Tracking the mind dur-
ing reading: The influence of past, present, and future words on fixation
durations. Journal of Experimental Psychology: General, 135, 12-35.
Koh, S., & Clifton, C. (2002). Resolution of the antecedent of a plural
pronoun: Ontological categories and predicate symmetry. Journal of
Memory & Language, 46, 830-844.
Kutas, M., & Federmeier, K. D. (1998). Minding the body. Psycho-
physiology, 35, 135-150.
814 Ledoux, Gordon, CambLin, and Swaab
Liversedge, S. P., Paterson, K. B., & Pickering, M. J. (1998). Eye
movements and measures of reading time. In G. Underwood (Ed.),
Eye guidance in reading and scene perception (pp. 55-75). Amster-
dam: Elsevier.
Liversedge, S. P., Pickering, M. J., Clayes, E. L., & Branigan, H. P.
(2003). Thematic processing of adjuncts: Evidence from an eye-tracking
experiment. Psychonomic Bulletin & Review, 10, 667-675.
Mimura, M., Verfaellie, M., & Milberg, W. P. (1997). Repetition
priming in an auditory lexical decision task: Effects of lexical status.
Memory & Cognition, 25, 819-825.
Moxey, L. M., Sanford, A. J., Sturt, P., & Morrow, L. I. (2004).
Constraints on the formation of plural reference objects: The influence
of role, conjunction, and type of description. Journal of Memory &
Language, 51, 346-364.
Nagy, M. E., & Rugg, M. D. (1989). Modulation of event-related poten-
tials by word repetition: The effects of inter-item lag. Psychophysiol-
ogy, 26, 431-436.
O’Brien, E. J., Albrecht, J. E., Hakala, C. M., & Rizzella, M. L.
(1995). Activation and suppression of antecedents during reinstate-
ment. Journal of Experimental Psychology: Learning, Memory, &
Cognition, 21, 626-634.
Osterhout, L., Allen, M., & McLaughlin, J. (2002). Words in the
brain: Lexical determinants of word-induced brain activity. Journal of
Neurolinguistics, 15, 171-187.
Paller, K. A., & Kutas, M. (1992). Brain potentials during memory re-
trieval provide neurophysiological support for the distinction between
conscious recollection and priming. Journal of Cognitive Neurosci-
ence, 4, 375-391.
Paller, K. A., Kutas, M., & McIsaac, H. K. (1995). Monitoring con-
scious recollection via the electrical activity of the brain. Psychologi-
cal Science, 6, 107-111.
Raney, G. E., Therriault, D. J., & Minkoff, S. R. B. (2000). Repeti-
tion effects from paraphrased text: Evidence for an integrated rep-
resentation model of text representation. Discourse Processes, 29,
61-81.
Rayner, K. (1978). Eye movements in reading and information process-
ing. Psychological Bulletin, 85, 618-660.
Rayner, K. (1998). Eye movements in reading and information process-
ing: 20 years of research. Psychological Bulletin, 124, 372-422.
Rugg, M. D. (1985). The effects of semantic priming and word repeti-
tion on event-related potentials. Psychophysiology, 22, 642-647.
Rugg, M. D. (1990). Event-related brain potentials dissociate repetition
effects of high- and low-frequency words. Memory & Cognition, 18,
367-379.
Rugg, M. D., Furda, J., & Lorist, M. (1988). The effects of task on
the modulation of event-related potentials by word repetition. Psycho-
physiology, 25, 55-63.
Sanford, A. J., & Garrod, S. C. (1981). Understanding written lan-
guage: Explorations of comprehension beyond the sentence. New
York: Wiley.
Sanford, A. J., & Garrod, S. C. (1998). The role of scenario mapping
in text comprehension. Discourse Processes, 26, 159-190.
Sanford, A. J., & Lockhart, F. (1990). Description types and method
of conjoining as factors influencing plural anaphora: A continuation
study of focus. Journal of Semantics, 7, 365-378.
Sanford, A. J., & Moxey, L. M. (1995). Notes on plural reference
and the scenario-mapping principle in comprehension. In C. Habel
& G. Rickheit (Eds.), Focus and coherence in discourse processing
(pp. 18-34). Berlin: de Gruyter.
Scarborough, D. L., Cortese, C., & Scarborough, H. S. (1977).
Frequency and repetition effects in lexical memory. Journal of Experi-
mental Psychology: Human Perception & Performance, 3, 1-17.
Sereno, S. C., & Rayner, K. (2003). Measuring word recognition in
reading: Eye movements and event-related potentials. Trends in Cog-
nitive Sciences, 7, 489-493.
Sereno, S. C., Rayner, K., & Posner, M. I. (1998). Establishing a
time-line of word recognition: Evidence from eye movements and
event-related potentials. NeuroReport, 9, 2195-2200.
Swaab, T. Y., Camblin, C. C., & Gordon, P. C. (2004). Reversed lexi-
cal repetition effects in language processing. Journal of Cognitive
Neuroscience, 16, 715-726.
Swick, D., & Knight, R. T. (1997). Event-related potentials differenti-
ate the effects of aging on word and nonword repetition in explicit and
implicit memory tasks. Journal of Experimental Psychology: Learn-
ing, Memory, & Cognition, 23, 123-142.
Traxler, M. J., Foss, D. J., Seely, R. E., Kaup, B., & Morris, R. K.
(2000). Priming in sentence processing: Intralexical spreading acti-
vation, schemas, and situation models. Journal of Psycholinguistic
Research, 29, 581-594.
van Berkum, J. J. A., Hagoort, P., & Brown, C. M. (1999). Semantic
integration in sentences and discourse: Evidence from the N400. Jour-
nal of Cognitive Neuroscience, 11, 657-671.
Van Petten, C., & Kutas, M. (1991). Influences of semantic and syn-
tactic context on open- and closed-class words. Memory & Cognition,
19, 95-112.
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, 131-150.
Yang, C. L., Gordon, P. C., Hendrick, R., & Hue, C. W. (2003). Con-
straining the comprehension of pronominal expressions in Chinese.
Cognition, 86, 283-315.
NOTES
1. The relation is called a c-command and is defined as follows: a
c-commands b if and only if the first branching node above a contains b
(see Gordon & Hendrick, 1997, 1998b, for a discussion and an alterna-
tive formulation of the critical syntactic relation).
2. Because of the presentation rate used in the experiment, the signal
to the critical word in the LPC time window overlaps with that of the
following word. There are two reasons to believe that the LPC effects we
see were due to processing of the critical word. First, the word following
the critical word was the same in all the experimental conditions; differ-
ences in processing this word were thus unlikely to arise. Second, if we
time-locked the signal to the following word, we did not see significant
differences in the ERP.
CoreferenCe and LexiCaL repetition 815
APPENDIX
Twenty Examples of the Experimental Stimuli in the Four Conditions Used in the Experiments
1. Last week Nicole (and Duncan) joined protests against the tuition hike because Nicole/Bonnie could not
afford the new rate.
2. Understandably Andre (and Molly) left the party early after Andre/Devin made a rude comment at dinner.
3. At the mall Gwendolyn (and Frederick) shopped for tents before Gwendolyn/Priscilla went camping.
4. Out in the field Jeffrey (and Serena) set up the telescope before Jeffrey/Anthony started looking at the
moon.
5. Yesterday Patricia (and Clifford) bought a new sweater because Patricia/Jennifer had spilled red wine all over
the old one.
6. With reluctance Dylan (and Edith) washed the dishes while Dylan/Wayne talked about the upcoming
election.
7. Despite the distance Sheila (and Eugene) looked for a house near the college after Sheila/Deidre was mugged
downtown.
8. Last Friday Kevin (and Doris) left work early after Kevin/Scott completed work on the project.
9. Based on the schedule Alexis (and Calvin) wrote the lyrics to the song before Alexis/Sandra composed the
music.
10. In spite of the rain Jared (and Norma) enjoyed the concert at which Jared/Damon met the band.
11. During the night Leila (and Bruce) called home because Leila/Dinah had crashed the car.
12. If asked Miguel (and Helene) always sang at parties when Miguel/Emmett rolled out the piano.
13. With great care Felicia (and Charles) painted the living room while Felicia/Dolores was on vacation from
work.
14. Quite spontaneously Edwin (and Shari) got married when Edwin/Oscar lived in Washington.
15. Despite the weather Janine (and Nathan) went to the beach when Janine/Violet rented a house for a week.
16. After the game Irving (and Gloria) got pizza because Irving/Norris doesn’t like Chinese food.
17. Until last year Cynthia (and Vincent) often bought books online until Cynthia/Natalie thought of borrowing
from the library.
18. Eventually Kenny (and Rosie) stopped playing golf because Kenny/Aaron said it was a waste of time.
19. Fearfully Ashley (and Roland) gasped in horror before Ashley/Daphne discovered that the horrible scars
were just make-up.
20. Every week Albert (and Hannah) went to the theater because Albert/Oliver gave free acting lessons.
(Manuscript received September 1, 2005;
revision accepted for publication March 20, 2006.)
... Once a referent has been activated, the antecedent representation can be integrated into the unfolding representation of the described event (Almor & Nair, 2007;Garrod & Sanford, 1994;Garrod & Terras, 2000;McKoon & Ratcliff, 1989;Nieuwland & Martin, 2017;Sanford et al., 1983;Sturt, 2003). Semantic integration takes place regardless of whether the expression is anaphoric or non-anaphoric, and is facilitated when the unfolding meaning of the sentence is coherent with respect to the preceding context and consistent with what can be expected based on world knowledge (e.g., Camblin, Gordon, & Swaab, 2007;Graesser, Millis, & Zwaan, 1997;Hagoort, Baggio, & Willems, 2009;Hagoort & Van Berkum, 2007;Kintsch & Van Dijk, 1978;Menenti, Petersson, Scheeringa, & Hagoort, 2008;Nieuwland & Van Berkum, 2006a;Van Berkum, Hagoort, & Brown, 1999;Zwaan & Rapp, 2006). ...
... In recognition memory studies, correctly recognized old words elicit smaller N400 amplitude than correctly rejected new words (Curran, 1999;Rugg et al., 1998;Van Strien, Hagenbeek, Stam, Rombouts, & Barkhof, 2005; for a review, see; Rugg & Curran, 2007). In studies on sentence and discourse comprehension, repeated referents (i.e., anaphora) elicit smaller N400 amplitude than new referents (Anderson & Holcomb, 2005;Camblin, Ledoux, Boudewyn, Gordon, & Swaab, 2007;Ledoux, Gordon, Camblin, & Swaab, 2007;Streb, Hennighausen, & R€ osler, 2004;Van Petten et al., 1991). Notably, the N400 component is not only sensitive to word repetition, but also to the fit between a word and the discourse context. ...
... Notably, the N400 component is not only sensitive to word repetition, but also to the fit between a word and the discourse context. Numerous studies have shown that discourse-coherent words elicited smaller N400 amplitude than discourse-incoherent words, and these discourse-level N400 effects appear identical or highly similar to the sentence-level N400 effect (Camblin, Gordon, et al., 2007;Filik & Leuthold, 2008;Nieuwland & Van Berkum, 2006a;Salmon & Pratt, 2002;St. George, Mannes, & Hoffinan, 1994;Van Berkum, Brown, & Hagoort, 1999;Van Berkum, Zwitserlood, Hagoort, & Brown, 2003). ...
... Once a referent has been activated, the antecedent representation can be integrated into the unfolding representation of the described event (Almor & Nair, 2007;Garrod & Sanford, 1994;Garrod & Terras, 2000;McKoon & Ratcliff, 1989;Nieuwland & Martin, 2017;Sanford et al., 1983;Sturt, 2003). Semantic integration takes place regardless of whether the expression is anaphoric or non-anaphoric, and is facilitated when the unfolding meaning of the sentence is coherent with respect to the preceding context and consistent with what can be expected based on world knowledge (e.g., Camblin, Gordon, & Swaab, 2007;Graesser, Millis, & Zwaan, 1997;Hagoort, Baggio, & Willems, 2009;Hagoort & Van Berkum, 2007;Kintsch & Van Dijk, 1978;Menenti, Petersson, Scheeringa, & Hagoort, 2008;Nieuwland & Van Berkum, 2006a;Van Berkum, Hagoort, & Brown, 1999;Zwaan & Rapp, 2006). ...
... In recognition memory studies, correctly recognized old words elicit smaller N400 amplitude than correctly rejected new words (Curran, 1999;Rugg et al., 1998;Van Strien, Hagenbeek, Stam, Rombouts, & Barkhof, 2005; for a review, see; Rugg & Curran, 2007). In studies on sentence and discourse comprehension, repeated referents (i.e., anaphora) elicit smaller N400 amplitude than new referents (Anderson & Holcomb, 2005;Camblin, Ledoux, Boudewyn, Gordon, & Swaab, 2007;Ledoux, Gordon, Camblin, & Swaab, 2007;Streb, Hennighausen, & R€ osler, 2004;Van Petten et al., 1991). Notably, the N400 component is not only sensitive to word repetition, but also to the fit between a word and the discourse context. ...
... Notably, the N400 component is not only sensitive to word repetition, but also to the fit between a word and the discourse context. Numerous studies have shown that discourse-coherent words elicited smaller N400 amplitude than discourse-incoherent words, and these discourse-level N400 effects appear identical or highly similar to the sentence-level N400 effect (Camblin, Gordon, et al., 2007;Filik & Leuthold, 2008;Nieuwland & Van Berkum, 2006a;Salmon & Pratt, 2002;St. George, Mannes, & Hoffinan, 1994;Van Berkum, Brown, & Hagoort, 1999;Van Berkum, Zwitserlood, Hagoort, & Brown, 2003). ...
Preprint
Full-text available
A key challenge in understanding stories and conversations is the comprehension of 'anaphora', words that refer back to previously mentioned words or concepts ('antecedents'). In psycholinguistic theories, anaphor comprehension involves the initial activation of the antecedent and its subsequent integration into the unfolding representation of the narrated event. A recent proposal suggests that these processes draw upon the brain's recognition memory and language networks, respectively, and may be dissociable in patterns of neural oscillatory synchronization (Nieuwland & Martin, 2017). We addressed this proposal in an electroencephalogram (EEG) study with pre-registered data acquisition and analyses, using event-related potentials (ERPs) and neural oscillations. Dutch participants read two-sentence mini stories containing proper names, which were repeated or new (ease of activation) and coherent or incoherent with the preceding discourse (ease of integration). Repeated names elicited lower N400 and Late Positive Component amplitude than new names, and also an increase in theta-band (4-7 Hz) synchronization, which was largest around 240-450 ms after name onset. Discourse-coherent proper names elicited an increase in gamma-band (60-80 Hz) synchronization compared to discourse-incoherent names. This effect was largest around 690-1000 ms after name onset and was localized to the left frontal cortex. We argue that the initial activation and subsequent discourse-level integration of referents can be dissociated with event-related EEG activity, and are associated with respectively theta- and gamma-band activity. These findings further establish the link between memory and language through neural oscillations.
... Kutas and Federmeier, 2011). In the discourse literature, N400 effects are, for example, reported in replications of the repeated name penalty using ERPs (Swaab et al., 2004;Camblin et al., 2007;Ledoux et al., 2007;Almor et al., 2017). Camblin et al. (2007), for instance, found increased N400 amplitudes following repeated name anaphors as compared to pronoun anaphors. ...
... The present results are compatible with previous experiments in so far as the literature on event-related potentials during referential processing consistently reports increased N400 amplitudes related to unpredicted referential relations based on prominence information (Swaab et al., 2004;Nieuwland and van Berkum, 2006;Camblin et al., 2007;Ledoux et al., 2007;Schumacher and Baumann, 2010;Hung and Schumacher, 2012;Schumacher and Hung, 2012;Wang and Schumacher, 2013;Almor et al., 2017). With the present analysis explicitly contrasting the referential form of anaphoric expressions and of their antecedents, we were able to show that prominence information based on referential form is already relevant for processing between 200 and 300 milliseconds after anaphor onset, and thus, earlier than the N400 time window usually considered crucial for referential processing. ...
Article
Full-text available
In discourse pragmatics, different referential forms are claimed to be indicative of the cognitive status of a referent in the current discourse. Referential expressions thereby possess a double function: They point back to an (existing) referent (form-to-function mapping), and they are used to derive predictions about a referent’s subsequent recurrence in discourse. Existing event-related potential (ERP) research has mainly focused on the form-to-function mapping of referential expression. In the present ERP study, we explore the relationship of form-to-function mapping and prediction derived from the antecedent of referential expressions in naturalistic auditory language comprehension. Specifically, the study investigates the relationship between the form of a referential expression (pronoun vs. noun) and the form of its antecedent (pronoun vs. noun); i.e., it examines the influence of the interplay of predictions derived from an antecedent (forward-looking function) and the form-to-function mapping of an anaphor (backward-looking function) on the ERPs time-locked to anaphoric expressions. The results in the time range of the P300 and N400 allow for a dissociation of these two functions during online language comprehension.
... On the other hand, a small number of ERP studies have provided neural information on pronoun resolution. Ledoux et al. suggested that the processing of pronoun resolution occurs between 250 and 500 milliseconds [25]. According to a review work of Nieuwland and Van Berkum, the processing of ambiguous pronoun resolution will induce a continuous frontal negative-going ERP [26]. ...
Article
Full-text available
Pronoun resolution plays an important role in language comprehension. However, little is known about its recruited cognitive mechanisms. Our investigation aims to explore the cognitive mechanisms underlying various types of pronoun resolution in Chinese using an electroencephalograph (EEG). We used three convolutional neural networks (CNNs)—LeNeT-5, GoogleNet, and EffifcientNet—to discover high-level feature abstractions of the EEG spatial topologies. The output of the three models was then fused using different scales by principal component analysis (PCA) to achieve cognitive workload classification. Overall, the workload classification rate by fusing three deep networks can be achieved at 55–63% in a participant-specific manner. We provide evidence that both the behavioral indicator of reaction time and the neural indicator of cognitive workload collected during pronoun resolution vary depending on the type of the pronoun. We observed an increase in reaction time accompanied by a decrease of the theta power while participants were processing ambiguous pronoun resolution compared to unambiguous controls. We propose that ambiguous pronoun resolution involves a more time-consuming yet more flexible cognitive mechanism, consistent with the predictions of the decision-making framework from an influential pragmatic tradition. Our results extend previous research that the cognitive states of resolving ambiguous and unambiguous pronouns are differentiated, indicating that cognitive workload evaluated using the method of machine learning for analysis of EEG signals acts as a complementary indicator for studying pronoun resolution and serves as an important inspiration for human–machine interaction.
... due to their 'earlier' access). On the other hand, the ability of context information to override other sources of facilitation on the N400 is seen not just for word frequency, but also for lexical association and word priming effects, which show a similar interaction with context effects as that seen here and previously for frequency (Camblin et al. 2007a;Ledoux et al. 2007;Van Petten & Kutas 1991). Kretzschmar et al. (2015) also examined N400 amplitudes in a fixation-related potential paradigm, using an extreme-group sample design, crossing items high and low in cloze probability and word frequency. ...
Article
In the current paper, we examined the effects of lexical (e.g. word frequency, orthographic neighborhood density) and contextual (e.g. word predictability in the form of cloze probability) features on single-trial event-related brain potentials in a self-paced reading paradigm. Critically, we examined whether individual differences in reading speed modulated single-trial effects on the N400, an ERP component linked to semantic memory access. Consistent with past work, we found that word frequency effects on the N400 were attenuated with increasing predictability. However, effects of orthographic neighborhood density were robust across all levels of predictability. Importantly, individual differences in reading speed moderated the influence of both frequency and predictability (but not orthographic neighborhood density) on the N400, such that slower readers showed reduced effects compared to faster readers. These data show that different lexical factors influence word processing through dissociable mechanisms. Our findings support a dynamic semantic-memory access model of the N400, in which information at multiple levels (lexical, sentential, individual) simultaneously contributes to the unfolding neural dynamics of comprehension.
... This result is consistent with the findings of Delogu et al. (2010) and suggests that processing the definite NP was facilitated when the noun could be interpreted as relevant to addressing the implicit QUD (Did John get a beer?). It could be argued, however, that rather than a facilitation for the nonactual context, the N400 effect at the noun reflects a "repeated NP penalty" (e.g., Almor, 1999) for the definite NP in the actual condition, where a pronoun would have been more felicitous (see, e.g., Ledoux, Gordon, Camblin, & Swaab, 2007;Swaab, Camblin, & Gordon, 2004, for repeated-name penalty N400 effects). According to theories that attempt to explain distributional patterns of referential forms in language use (e.g., Almor & Eimas, 2008;Almor, 1999Almor, , 2000Ariel, 1990;Gordon, Grosz, & Gilliom, 1993;Grosz, Joshi, & Weinstein, 1983), the more accessible a referent is, the less likely it is to be referred to by an explicit referring expression such as a (repeated) definite NP. ...
Article
Questions under Discussion (QUDs) have been suggested to influence the integration of individual utterances into a discourse-level representation. Previous work has shown that processing ungrammatical ellipses is facilitated when the elided material addresses an implicit QUD raised through a nonactuality implicature (NAIs). It is not clear, however, if QUDs influence discourse coherence during comprehension of fully acceptable discourse. We present two ERP studies examining the effects of QUDs introduced by NAIs using two-sentence discourses. Experiment 1 showed that processing definite NPs with inaccessible antecedents is facilitated when their content is relevant to the QUD. Using acceptable discourses, Experiment 2 showed that definite NPs failing to address a QUD elicit increased processing cost. Overall, our results indicate that QUDs raise the expectation that the following discourse will address them, providing unambiguous evidence that their influence is not limited to the processing of ungrammatical input.
... As a result, all RNP studies to date have employed reading tasks in which the dependent measure is whole-fragment or whole-sentence reading times instead of using a single word as a region of interest for analyses. Recent studies using the EEG methodology have demonstrated, however, that the RNP also takes the form of an increased N400 amplitude triggered by the repeated anaphor (Huang et al., 2014;Ledoux et al., 2007;Swaab et al., 2004). This is an important finding because the N400 is taken to be an index of immediate processing difficulty (Kutas & Hillyard, 1984): Given that most theories of reference processing hypothesise that the effect of referential form is manifested at later integrative stages during reading (when the reference is resolved and the discourse representation is updated), it is remarkable that EEG studies found evidence for RNP effects 400 milliseconds after the onset of the critical anaphor. ...
Preprint
Full-text available
Investigations of coreferential processing typically require participants to link anaphors with semantically underspecified (“empty”) discourse entities. However, outside the laboratory, anaphors often refer to people, objects, or events about which we possess extensive background knowledge. In addition, recent evidence indicates that comprehenders experience processing difficulty when sentence characters are semantically similar. In the current study we examined whether activating pre-existing real-world knowledge about antecedents influenced coreferential processing in a developing sentence context. Event-related potentials were recorded as participants read sentences containing ambiguous pronouns. Antecedents were either “empty” or were real, well-known individuals. In addition, pronouns either matched or mismatched the sex of their antecedents. Mismatching anaphors elicited a P600 effect whose amplitude was significantly greater when sentence characters were real. Moreover, matching pronouns elicited a P600-like effect when their antecedents were semantically “empty”. Our results suggest that the presence of high-quality representations in a discourse model facilitates coreferential processing.
Preprint
Full-text available
Establishing referential relations among text elements is critical to establishing discourse coherence. Although resolution of coreference between anaphors and their antecedents is central to many theories of text processing, most address the semantic features of antecedent entities in a minimal way (e.g., via morphological features such as gender and number). However, research shows that semantically rich entities are represented in a different manner than semantically empty ones (e.g., Bill Clinton versus Bill Smith), and that such differences have processing consequences in a variety of cognitive tasks. In two experiments, we investigated how the semantic features of discourse referents affected coreferential processing. Our results indicate that increasing the semantic detail associated with the characters in our experimental sentences increased their psychological prominence in the discourse representation. In Experiment 1, semantic detail increased the relative availability of antecedent entities in memory when they were uniquely identifiable by an anaphoric expression. In Experiment 2, semantic detail also affected the online resolution and integration of anaphor – antecedent relations: semantically rich antecedents that were not otherwise focused in the discourse reliably elicited the repeated-name penalty, an effect known to reflect discourse prominence. In addition, semantic detail seemed to permit recovery from lingering effects of the repeated-name penalty. Our results are consistent with recent evidence that the quality of discourse information affects the construction of a coherent representation, even when it is incidental to structural and referential relations in a text. Models of text processing must incorporate a role for this kind of representational information.
Chapter
This chapter illustrates the use of reading time measures. The time measures sum fixation durations to gain an understanding of the difficulty experienced while reading texts. A distinction is made between two approaches to sum the duration of fixations. The first approach is to sum the duration of fixations which neighbor each other in a specified region of space. The second approach is to sum fixations that occur in a sequence over a specified period of time. It is argued that both types of reading time measures are needed if the experimenter is to understand the time course of the influence of a linguistic variable on readers' processing of text. Primarily, a number of hypothetical eye movement records are discussed to illustrate the differential sensitivities of different eye movement measures. An eye movement experiment investigates how people process reduced relative clause sentences with and without the focus operator only to examine the utility of different reading time measures is recorded. The results show that measures summing temporally contiguous fixations can make an important contribution to the experimenters' understanding of the precise pattern of eye movements which occur when a problem is encountered in the text.
Article
THE average duration of eye fixations in reading places constraints on the time for lexical processing. Data from event related potential (ERP) studies of word recognition can illuminate stages of processing within a single fixation on a word. In the present study, high and low frequency regular and exception words were used as targets in an eye movement reading experiment and a high-density electrode ERP lexical decision experiment. Effects of lexicality (word vs pseudoword vs consonant strings), word frequency (high vs low frequency) and word regularity (regular vs exception spelling-sound correspondence) were examined. Results suggest a very early time-course for these aspects of lexical processing within the context of a single eye fixation.
Article
Event-related potentials (ERPs) were recorded during the performance of a lexical decision task, in which a proportion of the words were either semantic associates or repetitions of the preceding word. Reaction times were faster to both the second member of associated pairs (targets) and repeated words, with the latter facilitatory effect being more than twice that of the former.ERPs to the semantic primes were more negative-going than those to targets. This difference peaked around 400–450 ms after stimulus onset. Comparison of unrepeated and repeated words revealed a larger and temporally more extended difference, with a similar scalp topography.The prime-target differences are interpreted as a further example of the sensitivity of the ‘N400’ component of the ERP to semantic relationships between words. The differences between the repetition and priming effects in ERPs are considered equivocal with respect to the view that the associated behavioural effects are caused by different cognitive mechanisms.
Article
Abstract The N400 is an endogenous event-related brain potential (ERP) that is sensitive to semantic processes during language comprehension. The general question we address in this paper is which aspects of the comprehension process are manifest in the N400. The focus is on the sensitivity of the N400 to the automatic process of lexical access, or to the controlled process of lexical integration. The former process is the reflex-like and effortless behavior of computing a form representation of the linguistic signal, and of mapping this representation onto corresponding entries in the mental lexicon. The latter process concerns the integration of a spoken or written word into a higher-order meaning representation of the context within which it occurs. ERPs and reaction times (RTs) were acquired to target words preceded by semantically related and unrelated prime words. The semantic relationship between a prime and its target has been shown to modulate the amplitude of the N400 to the target. This modulation can arise from lexical access processes, reflecting the automatic spread of activation between words related in meaning in the mental lexicon. Alternatively, the N400 effect can arise from lexical integration processes, reflecting the relative ease of meaning integration between the prime and the target. To assess the impact of automatic lexical access processes on the N400, we compared the effect of masked and unmasked presentations of a prime on the N400 to a following target. Masking prevents perceptual identification, and as such