INTRODUCTION
Semantic priming refers to the facilitation of target information when
it is preceded by related information, presumably due to the proximity of
associated concepts in semantic memory. There is a large body of
literature implicating several different brain regions in semantic priming
(Copland et al., 2003; Kotz
et al., 2002; Rissman et al., 2003; Rossell et al., 2003; Seger et al.,
2000; for a review, see Henson, 2003),
but one emerging idea is that the left prefrontal cortex plays an
important role in this process due to its proposed role in the retrieval
of semantic information (Buckner et al., 1995;
Demb et al., 1995; Gabrieli
et al., 1996; Kapur et al., 1994; Tulving et al., 1994; Wagner et
al., 2001), use of contextual cues and source information to
facilitate semantic processing (Janowsky et al.,
1989; Shimamura et al., 1990), and/or
selection of target information among competing alternatives in semantic
memory (Thompson-Schill et al., 1997, 1999).
There is also a growing body of literature to suggest that the right
prefrontal cortex contributes to semantic priming, although its exact role
in priming has not been fully elaborated. Recent neuroimaging and
event-related potentials (ERP) studies provide some evidence that the
right frontal lobe is involved in processing both typical and atypical
semantic relationships (i.e., noun–verb relations), whereas the left
frontal lobe is only adept at processing typical semantic associations
(Beeman, 1998; Kiefer et al.,
1998; Seger et al., 2000). Divided visual
field studies implicate the right hemisphere in the processing of remote
associations and indicate that a wider range of word meanings is
represented in the right hemisphere as compared to the left (Chiarello & Richards, 1992; Coney & Evans, 2000; Faust
& Chiarello, 1998; Nakagawa, 1991;
Titone, 1998). In addition, evidence from lesion
studies suggests that the right frontal lobe contributes to discourse
comprehension and interpreting metaphorical expressions (Brownell, 1988), as well as in the use and
appreciation of humor (Shammi & Stuss,
1999). Together, these studies suggest that the right prefrontal
cortex is involved in several aspects of semantic processing that may
contribute to semantic priming effects and appear to be qualitatively
different from those afforded to the left frontal lobe.
There are several methods that can be used to evaluate semantic
processes, and one well-established method is to examine priming effects
when participants use contextual cues to resolve lexical ambiguities in
speech or text (Copland et al., 2000a). To date,
most studies of lexical ambiguity resolution in patient populations have
focused exclusively on patients with left hemisphere damage (Blumstein et al., 1982; Grinrod
& Baum, 2002; Hagoort, 1993; Milberg & Blumstein, 1981; Milberg et al., 1987) or subcortical lesions (Copland
et al., 2000a, 2000b).
However, there is some recent patient data to suggest that right
hemisphere damage also disrupts lexical-semantic priming. In a series of
studies by Tompkins and colleagues, patients with right hemisphere damage
showed difficulty suppressing contextually inappropriate word meanings in
sentence comprehension relative to a group of healthy controls (Tompkins
et al., 1997, 2000,
2001). In these studies, however, damage to the
right hemisphere was diffuse and the degree to which the right frontal
lobe contributed uniquely to this effect could not be determined.
We are aware of only one study that has examined semantic priming
using a lexical-ambiguity resolution task in patients with discrete right
and left frontal lobe lesions. Metzler (2001)
studied lexical-ambiguity resolution in patients with right, left, and
bilateral frontal lobe lesions on a word-reading task. In her study,
patients were presented with word triplets in which the second word in
each triplet was always a homograph and the first and third words were
either related to the same (concordant) or different (discordant) meanings
of the homograph. Whereas the former condition produces semantic
facilitation (i.e., faster reaction times to the target word), the latter
condition does not produce facilitation (i.e., slower reaction times to
the target) in young healthy participants. Using proportional reaction
time scores (i.e., discordant–concordant/concordant), Metzler
found that semantic priming for concordant meanings of homographs was
abolished in patients with left and bilateral frontal lobe lesions
relative to discordant meanings, whereas semantic priming for concordant
meanings was facilitated in patients with right frontal lobe lesions
relative to those with left frontal lesions and healthy controls. Although
admittedly post hoc, Metzler proposed that the increased
facilitation of patients with right frontal lobe damage in the concordant
condition may reflect an overreliance on responses that are prepotent and
expected, and that this may explain the role of the right frontal lobe in
perseveration (Lombardi et al., 1999).
Although the study by Metzler helped to provide important insights
into the role of the right frontal lobe in semantic priming, there are two
potential limitations of that study that make it difficult to determine
the exact nature of semantic-priming deficits in patients with right
frontal lobe lesions. One limitation is that only patients with right
medial and/or bifrontal lesions were included in the study, and in
several cases the location of the lesion was uncertain (see Metzler, 2001; Table 1).
Thus, it is unclear whether patients with lesions that extend beyond the
right medial frontal lobe produce the same or a different pattern of
semantic priming. Second, an experimental baseline condition was not
included in the study by Metzler, but rather priming was defined as the
proportional facilitation of the concordant relative to the discordant
condition. Without a baseline condition, it is difficult to determine
exactly which condition (i.e., concordant or discordant) is abnormal in
patients with right frontal lobe lesions.
Patient group, etiology, surgery location, demographic information,
and neuropsychological performances for individual patients
In the present study, we further examined semantic priming in patients
with right frontal lobe lesions using a lexical-ambiguity resolution task
very similar to the one used by Metzler. However, in our study, we
addressed the limitations of previous work in this area by examining a
group of patients with more diverse lesion locations and by including two
additional experimental conditions in our task. Exploring regional
differences in semantic priming within the right frontal lobe is important
because there is a vast literature indicating that the frontal lobe is not
a unitary region, but rather contains many subregions that subserve
different cognitive functions (e.g., Gabrieli et al.,
1998). In particular, neuroimaging studies that have found right
frontal activation in semantic tasks have identified region specific
differences (e.g., Seger et al., 2000).
Therefore, in the present study, we examined a group of patients similar
to Metzler's who had lesions restricted to the right medial frontal
lobe, as well as a group of patients with lesions that included the right
dorsolateral region. The latter group was of interest because the right
dorsolateral frontal cortex has been implicated in other types of priming
(i.e., negative priming in selective attention; see Stuss et al., 2002), and we wished to determine
whether or not this region is also involved in semantic priming. In
addition, we included a baseline condition in which the three words were
unrelated (i.e., the unrelated condition), as well as a condition in which
there was no context to influence meaning selection (i.e., the neutral
condition). These conditions allowed us to isolate facilitation effects
within each group in order to clarify the exact nature of semantic-priming
deficits in patients with right medial and dorsolateral frontal lobe
lesions.
Based on previous literature on lexical-ambiguity resolution (Copland et al., 2000a; Hagoort,
1989), we hypothesized that control participants would show faster
responses in the concordant condition relative to all other conditions,
since in the concordant condition, attention is directed toward the
context-appropriate meaning of the word. In addition, we predicted that
controls would show slower responses in the discordant condition relative
to the neutral condition and either slower or equivalent responses in the
discordant condition relative to the unrelated condition. This pattern of
responses would provide strong evidence that context is influencing
meaning selection by facilitating context-appropriate meanings (i.e.,
concordant) but not context-inappropriate responses (i.e., discordant) or
irrelevant (i.e., unrelated) meanings. Based on Metzler's study, we
hypothesized that patients with damage to the right medial (RM) frontal
lobe would show increased facilitation (i.e., a hyperpriming effect), as
evidenced by significantly faster responses in the concordant condition
relative to the baseline conditions, and that facilitation in the
concordant condition would significantly exceed that of the controls. We
also hypothesized that due to the functional heterogeneity within the
frontal lobes, lesions involving the right dorsolateral cortex would
produce a different pattern of semantic priming than patients with lesions
circumscribed to the RM frontal lobes. This hypothesis was based on
existing research implicating the right dorsolateral cortex in the ability
to abolish or filter irrelevant information (Braver et
al., 2001; Perlstein et al., 2003; Stuss et al., 2002). Therefore, we hypothesized that
patients with involvement of the right dorsolateral frontal lobe may show
disrupted priming characterized by an inability to abolish priming in the
discordant condition.
METHODS
Participants
Participants in this investigation included 13 patients (22 to 51
years of age) who underwent a right frontal lobe resection for treatment
of intractable epilepsy. All patients were recruited from the University
of Florida Comprehensive Epilepsy Program (UFCEP) and were included in the
investigation if they had evidence of unilateral, focal frontal lobe
damage and a documented Full Scale IQ (FSIQ) above 70. Precise lesion
location was based on the operative reports and determined by the
neurosurgeon that performed the resections (S.N.R.) and a second
neurosurgeon that assisted with the lesion descriptions (R.J.B.). The
sample of participants consisted of six patients with RM frontal lobe
resections and seven with right extramedial frontal (RDL/M) lobe
resections (three dorsolateral, four dorsolateral plus medial). The
patient sample consisted of eight females and five males. All participants
were left-hemisphere dominant for language as determined by pre-surgical
Wada evaluations. Table 1 displays the seizure
etiology, surgical lesion location, gender, age, education level, and
neuropsychological data for each individual patient who received a right
frontal resection.
Twenty age- and education-matched healthy participants were recruited
as the control group. Fifteen were female and five were male. Table 2 provides the mean demographic characteristics,
epilepsy information, and neuropsychological data for the RM and RDL/M
groups, and demographic characteristics and neuropsychological data for
the control group. One-way analysis of variances (ANOVAs) revealed no
significant differences among the three groups in age or in level of
education (see Table 2). A 3 × 2
chi-squared Test did not reveal any differences among the three groups in
gender distribution, χ2 = 1.4, p = .502.
Independent t tests were also conducted between the RM and
RDL/M groups and revealed no reliable differences in age of seizure
onset, seizure duration, or years since surgery.
Demographic characteristics, epilepsy features, and neuropsychological
performances for of the RM, RDL/M, and Control groups (standard
deviations are in parentheses)
To better characterize the patients with RM and RDL/M lesions
cognitively, a short battery of neuropsychological tests was administered,
which included the Digit Span subtest of the Wechsler Memory
Scale–III (Wechsler, 1997), the Reading
subtest of the Wide Range Achievement Test–Third Edition
(WRAT–3; Wilkinson, 1993), the Boston
Naming Test (BNT; Kaplan et al., 1983), and the
Stroop Color-Word Test (Golden, 1978). The
normal controls were also administered a subset of these tests. One-way
ANOVAs did not reveal any significant differences among the three groups
on the Digit Span Test, WRAT standard score or the Interference index from
the Stroop Test. T tests indicated that the RM and the RDL/M
groups did not differ on full scale IQ (FSIQ), performance IQ (PIQ),
verbal IQ (VIQ), or the BNT. In general, these comparisons indicated that
the RM and RDL/M groups were not demonstrating significant
neuropsychological deficits and they did not differ from one another.
Further, neither performance on the Word Picture Matching subtest of the
Western Aphasia Battery nor behavioral observations made during the time
of testing revealed any obvious aphasia in any of the patients.
Stimuli
A modified version of the word triplet semantic-priming paradigm
described by Copland et al. (2000a) was used as
a measure of lexical-ambiguity resolution. Stimuli in this experiment
consisted of word triplets presented visually on a computer screen, in
which the first word was a word providing the context (context prime), the
second word represented an ambiguous prime (homograph), and the third word
was the target word. Stimuli consisted of 12 lexically ambiguous balanced
homographs (i.e., bank) that had two unrelated, distinct, and common
meanings for which there are relatively strong associates (Twilley et al., 1994). Four word triplets were
constructed for each lexical ambiguity and presented within the context of
four different experimental conditions—concordant,
discordant, neutral, and unrelated conditions
(see Table 3). In the concordant
condition, the context prime and the target were related to the same
meaning of the ambiguous prime. In the discordant condition, the
context prime and the target were related to different meanings of the
ambiguous prime. In the neutral condition, the context prime was
unrelated to the meaning of the ambiguous prime indicated by the target.
In the unrelated (baseline) condition, the target remained the
same, while the two primes were words unrelated to each other or to the
target. While the concordant and discordant conditions evaluate the
effects of a preceding context on target performance, the neutral
condition allowed us to examine the effects of target performance in the
absence of contextual constraints (i.e., neither meaning of the homograph
is primed). Previous studies using a neutral condition have shown that
this condition produces less priming than the concordant condition, but
that some priming is observed as responses are typically faster in the
neutral condition compared to the discordant and unrelated conditions
(Hagoort, 1989). Therefore, response times in
the discordant and concordant conditions are directly compared to the
unrelated and neutral conditions when determining the effects of context
on meaning selection.
Four experimental conditions in the Word Triplets Task
Each critical target was presented four times (once for each
condition) at each of two interstimulus intervals (ISIs; 100 ms and 1250
ms) in the experiment. Two ISIs were included in this experiment since
there is data to suggest that lexical-semantic priming effects at short
ISIs (i.e., 100 ms) may reflect automatic processes, whereas priming
effects at longer ISIs (i.e., 1250 ms) reflect controlled, attention-based
processes that are more susceptible to contextual constraints (Simpson, 1984). For both ISIs, the order of
presentation for each triplet with the same target was counterbalanced to
reduce repetition effects. In addition, no experimental condition occurred
more than three times in a row throughout the experiment and there were a
minimum of four intervening trials between triplets with the same target.
Twelve word sequences were also created in which the target stimulus is a
set of three question marks instead of a word. On these trials,
participants were required to recall the first two words on the trial. The
“recall” triplets were presented twice in a session and were
included to insure that participants attended to all stimuli. Overall, 60
triplets (48 critical and 12 recall) were presented to the participant. An
initial block of 10 practice trials was completed prior to testing. During
the practice trials, the sensitivity of the relay was adjusted to each
participant's voice.
Procedure
Participants were tested individually and seated in front of a
computer screen. They were told that they would see three stimuli
presented in succession in the center of the screen and that the third
stimulus would either be a word or set of three question marks. They were
instructed that if the third stimulus was a word, they should read the
word aloud as quickly and as accurately as possible. On the other hand, if
the third stimulus was a set of three question marks, they should quickly
try to recall the first two words presented on that trial. A single trial
consisted of a ready signal followed by a fixation cross which was
presented in the center of the screen for 200 ms, followed by the context
prime, the ambiguous prime, and the target presented for 200 ms each in
succession. In one block of trials, an ISI of 100 ms was placed between
the words in each triplet, while in another block of trials, an ISI of
1250 ms was placed between the word triplets. The two blocks were
counterbalanced across participants so that some received the 100-ms ISI
first and others received the 1250-ms ISI first. In both blocks,
participants initiated each trial by pressing the space bar. Reaction time
(RT) to the target word was measured and was recorded from the onset of
the target word to the onset of the participant's vocal response on
all trials. Errors were recorded by the examiner on each trial.
RESULTS
The dependent variable in this study was the vocal RT to the target
word. Statistical analyses were carried out on RTs to the target after
outliers (RTs > 2 SD from the mean) and errors (incorrect responses and
microphone failures) were removed. Removal of outliers resulted in the
exclusion of 8.3% of the controls' data, 9.4% of the RM groups'
data, and 8.5% of the RDL/M groups' data. Because errors
accounted for less than 1% of the data in each group, error analyses were
not conducted. Median RTs for each group across the four conditions are
displayed in Table 4. RTs for each trial were
normalized by log transformations in order to help improve variance
homogeneity. Prior to computing these difference scores, a one-way ANOVA
was conducted on the three groups' log transformed RTs in the
baseline (unrelated) condition in order to make sure there were no
baseline differences in RTs among the groups. This analysis did not reveal
any significant differences among the three groups, F(2, 38) =
1.3, p = .057, although the test approached significance.
Average median reaction times (RT) in msec and standard
deviations (SD) for the RM, RDL/M, and Control groups in the
four experimental conditions
Semantic-priming effects were examined by computing difference scores
between the log-transformed RTs from the unrelated condition and the
log-transformed RTs from the concordant, discordant, and neutral
conditions. An initial Group × ISI × Condition
repeated-measures ANOVA revealed a main effect for condition,
F(2, 60) = 3.46, p = .031, but no main effect for group,
F(2, 30) = 2.06, p = .145, or ISI, F(1, 30) =
3.8, p = .061. The main effect for condition was mediated by a
significant Group × Condition interaction, F(4, 60) = 5.17,
p = .001, indicating that the three groups demonstrated different
patterns of semantic priming across the experimental conditions. There was
no Group × ISI interaction, F(2, 30) = .61, p =
.524, nor was there a Condition × ISI interaction, F(2, 60)
= 2.51, p = .112. In addition, the Group × Condition
× ISI interaction was not significant, F(4, 60) = .28,
p = .887. Because the analyses did not reveal any main effect or
interactions with ISI, we examined group differences in the various
conditions collapsed across ISI using one-way ANOVAs with group (controls
vs. RM vs. RDL/M) as the factor.
Discordant Condition
Performance in the discordant condition was defined as the average
log-transformed RT in the unrelated condition minus the average
log-transformed RT in the discordant condition. Thus, a small or negative
value indicates the expected slowing in the discordant condition relative
to baseline. A one-way ANOVA in the discordant condition revealed
significant differences among the three groups, F(2, 32) = 6.45,
p = .005. Follow-up comparisons using Tukey's HSD tests
indicated that the control group was significantly different from the
RDL/M group, p = .002, and that the RDL/M group was
significantly different from the RM group, p = .012. The controls
were not significantly different from the RM group, p = .692. As
can be seen in the figure, the control group and the RM group showed
significantly slower performances in the discordant condition relative to
the RDL/M group. In contrast, the RDL/M group demonstrated faster
performances in the discordant condition relative to the baseline
condition, suggesting priming of the contextually inappropriate word
meaning.
To determine if the finding in either patient group was due to a
subset of participants with highly deviant scores, we examined each
participant's score in the discordant condition. In the control
group, 14/20 of the patients demonstrated the typical pattern, that
is, slower responses to the discordant meaning relative to baseline.
Similarly, 5/6 of the patients in the RM group showed this pattern. In
contrast, all 7 of the RDL/M patients displayed faster responses
(i.e., facilitation) to the discordant meaning. Furthermore, there was no
difference in the degree of facilitation between patients with pure
dorsolateral damage and those with dorsolateral plus medial damage,
t(5) = 1.25, p = .265, indicating that the group
findings were not due to a subset of patients.
Concordant Condition
Priming for the concordant meaning was defined as the average
log-transformed RT in the unrelated condition minus average
log-transformed RT in the concordant condition. Thus, a greater positive
value means greater facilitation of the concordant meaning of the words.
Results of a one-way ANOVA indicated that the three groups differed
reliably on this index, F(2, 32) = 4.8, p = .016.
Tukey's HSD tests indicated that both the control and RDL/M
groups were significantly different from the RM group, p's =
.025 and .004, respectively. The control and the RDL/M groups did not
differ significantly from one another, p = .616. As can be seen
in the figure, the control group and the RDL/M group demonstrated the
typical facilitation effect for the concordant meaning (i.e., faster
responding in the concordant relative to baseline condition). In contrast,
the RM group did not demonstrate facilitation in the concordant condition,
and obtained a slightly negative score. Individual participant analyses
revealed that 15/20 of the controls, 2/6 of the RM, and all 7 of
the RDL/M patients demonstrated facilitation in the concordant
condition.
Neutral Condition
To determine group differences in priming when contextual constraints
are not imposed, a neutral condition was included. A one-way ANOVA
(unrelated condition–neutral condition) indicated no group
differences in priming for targets presented in a neutral context,
F(2, 32) = 0.51, p = .603. As can be seen in the figure,
all three groups had positive scores in the neutral condition. Analysis of
individual participants revealed that 13/20 of the controls, 3/6
of the RM patients, and 4/7 of the RDL/M patients (2/4
“mixed” and 2/3 “pure dorsolateral”) showed
facilitation in the neutral condition.
Within-Group Comparisons
Because the control and patient groups showed different patterns of
priming across the experimental conditions, we examined priming effects
within each group in order to determine how context influences meaning
selection. To determine whether or not each group demonstrated
facilitation of the context-appropriate meaning, RTs in the concordant
condition were compared to RTs in the unrelated, discordant, and neutral
conditions. Planned pairwise comparisons revealed that the control group
displayed faster responses to the concordant condition relative to the
discordant, t(19) = 3.72, p = .001, and unrelated,
t(19) = 3.72, p = .001 conditions, whereas facilitation
of the concordant relative to the neutral condition only approached
significance, t(19) = 2.70, p = .068. This pattern
suggests that the control group demonstrated the expected facilitation of
the context-appropriate meaning, although facilitation was not
significantly greater than when no context was provided. The
RDL/M group demonstrated facilitation of the concordant meaning
relative to the unrelated t(6) = 7.78, p = .000, and
neutral, t(6) = 2.55, p = .050, conditions, but they did
not show greater facilitation of the concordant meaning relative to the
discordant meaning, t(6) = 0.03, p = .976. This pattern
suggests that while RDL/M group do show facilitation effects for the
context-appropriate meaning, facilitation does not appear to be greater
than that displayed for the inappropriate meaning. Unlike the control and
RDL/M groups, the RM group did not demonstrate facilitation of the
concordant meaning relative to the discordant, t(5) = 2.13,
p = .086, neutral, t(5) = 1.95, p = .108, or
unrelated, t(5) = .75, p = .486, conditions.
To establish whether or not each group demonstrated the expected lack
of facilitation for the context-inappropriate meaning, RTs in the
discordant condition were compared to RTs in the neutral and unrelated
conditions. Planned pairwise comparisons revealed that the control group
was significantly slower in the discordant condition relative to the
neutral condition, t(19) = 2.95, p = .008, suggesting
that the context-inappropriate meaning received less facilitation than the
condition in which there was no context to guide meaning selection. In
addition, there was no difference between the discordant and unrelated
conditions t(19) = 1.00, p = .522, suggesting that the
discordant meaning was not facilitated relative to baseline. The RM group
showed a similar pattern to the controls in that their discordant
condition was slower than the neutral, t(5) = 3.7, p =
.013, but no different than the unrelated, t(5) = 1.0, p
= .365, condition. Conversely, the RDL/M group showed greater
facilitation in the discordant condition relative to both the neutral,
t(6) = 2.43, p = .036, and unrelated, t(6) =
4.04, p = .007, conditions. Inspection of Figure 1 reveals that the RDL/M group displayed a
level of facilitation for the discordant meaning that was similar to that
seen for the concordant meaning. These results suggest that for the
RDL/M group, context does not influence meaning selection, and
therefore, contextually appropriate and inappropriate meanings both
receive facilitation.
Log transformed difference scores (unrelated condition −
experimental condition) for the Controls, RM, and RDL/M
groups.
Repetition Effects
Since previous research has suggested that target repetition can
differentially affect priming in different experimental conditions (Hagoort, 1989; but also see Milberg
& Blumstein's response, 1989), we tested whether or not
target repetition impacted priming and/or whether or not the effects
of repetition on priming differed among the groups in a Group ×
Repetition × Condition repeated-measures ANOVA. Results of this
analysis revealed a main effect of repetition, F(7, 210) = 3.14,
p = .004, a main effect of group, F(2, 30) = 5.3,
p = .011, and a main effect of condition, F(2,
60)=13.99, p = .000. As expected, based on our previous analysis,
these effects were mediated by a Group × Condition interaction,
F(4, 60) = 5.7, p = .001. Importantly, group did not
interact with repetition, F(10, 210) = .745, p = .681,
and there was no Group × Repetition × Condition interaction,
F(30, 450) = 1.2, p = .174, suggesting that the effects
of repetition on priming were uniform across the groups. There was,
however, a Repetition × Condition interaction, F(14, 420) =
1.26, p = .034. To further examine the different effects of
repetition on condition, we conducted paired t tests on the
difference scores between the first and last presentation of each target
in each condition. These results indicated that the unrelated condition
showed a greater effect of repetition (i.e., greater reduction in RTs)
than the neutral, t(32) = 2.65, p = .012, and
concordant, t(32) = 2.01, p = .049, conditions, but not
the discordant condition, t(32) = 1.9, p = .096. Taken
together, while target repetition had a significant effect on priming of
unrelated word targets, repetition effects cannot account for our group
differences in priming.
Comparison to Previous Findings
To provide a more direct comparison between our results and those
obtained by Metzler, we reanalyzed our data using the proportional
facilitation scores used by Metzler (i.e., [(discordant raw RT
− concordant raw RT)/concordant raw RT]). Using
proportional RT scores, Metzler's right (medial) frontal group showed
a semantic facilitation effect of .060 and her control group showed a
facilitation effect of .020. Conversely, our RM group showed a semantic
facilitation effect of only .031 and our control group showed semantic
facilitation effect of .048. Thus, our RM group did not demonstrate the
abnormal “hyperpriming” effect that Metzler reported, but
rather, the RM patients in the present study look more like the controls
when proportional priming scores are used.
Lesion Analysis
In this study, we compared a group of patients with pure RM lesions
(N = 6) to those with lesions extending beyond the RM cortex
(N = 7; right dorsolateral, right medial plus dorsolateral) so
that we could compare our RM group directly to Metzler's RM
group.a
A group of six patients with left
medial frontal damage were also studied using this paradigm and showed
facilitation of the concordant and discordant meanings at both
ISIs—a pattern of performances similar to our RDL/M group. This
finding is consistent with lesions studies and the fMRI literature
implicating the left frontal lobe in lexical-semantic priming (Metzler, 2001). However, we wished to focus the
current paper on the role of the right frontal cortex in semantic priming
since there is little information available on this topic, and therefore,
we do not present data on our patients with left frontal lobe lesions in
this manuscript. A full description of their performances is reported
elsewhere (McDonald et al., 2003).
However, we also wished to determine whether having any damage to the
right medial frontal lobe impacted performances. Therefore, we compared
patients with lesions that involved the RM frontal lobe (
N = 10;
RM plus the “mixed” medial/dorsolateral group) to those
with pure dorsolateral lesions (
N = 3). Independent
t
tests did not reveal any differences in the concordant,
t(11) =
1.5,
p = .139, discordant,
t(11) = .545,
p =
.597, or neutral,
t(11) = .162,
p = .874, priming
conditions between these groups. In addition, to provide a comparison of
patients with discrete, nonoverlapping lesions, we compared the RM
patients to those with pure right dorsolateral damage. Independent
t tests revealed that the two groups differed in their degree of
facilitation in the concordant condition,
t(7) = 2.67,
p
= .032. The pure dorsolateral group showed greater facilitation of the
concordant meaning relative to the RM group. Their priming scores did not
differ in the neutral condition,
t(7) = .330,
p =.751.
Group differences in the discordant priming condition approached
significance,
t(7) = 1.88,
p = .062. While the latter
comparison was not statistically significant, 5/6 of the patients in
the RM group showed slowing in the discordant condition, whereas none of
the pure RDL patients showed slowing in this condition. Therefore, it may
be that involvement of the right dorsolateral frontal cortex, either
exclusively or in combination with medial frontal damage, disrupts the
ability of context to guide meaning selection.
DISCUSSION
This study addressed lexical-ambiguity resolution in patients with RM
and RDL/M frontal lobe lesions and controls in order to determine
whether differences in semantic priming could be identified within regions
of the right hemisphere. Results indicated that whereas healthy controls
and patients with RM frontal lesions did not show facilitation of the
contextually inappropriate meaning of homographs, all 7 patients with
lesions that included the right dorsolateral cortex demonstrated
facilitation of the contextually inappropriate meaning relative to a
baseline condition. This finding suggests that lesions including the right
dorsolateral frontal lobe may disrupt selective access to the appropriate
homograph meaning, and instead allow for activation of multiple meanings
within a single context. Thus, we concluded that regional differences in
semantic priming do appear to exist within the right frontal lobe, at
least for the processing of lexically ambiguous information.
The nature of semantic priming within each patient group is complex
and should be further elaborated. First, the control group showed a
pattern of performances suggesting that they possess selective access to
context-appropriate meanings during lexical-ambiguity resolution. That is,
they showed facilitation of the concordant condition relative to baseline
and discordant conditions and facilitation of the neutral condition
relative to the discordant. They also showed a trend toward
facilitation of the concordant relative to the neutral condition. In
addition, they did not show facilitation of the discordant meaning. It is
of note that their performances were not significantly slower in the
discordant relative to the baseline condition, suggesting that the
discordant meaning was not necessarily inhibited, but rather, was
unaffected by the context prime. Overall, this pattern of performances is
consistent with previous research of lexical-ambiguity resolution in
healthy controls suggesting that context influences meaning selection
(Grinrod & Baum, 2002; Hagoort, 1993).
Conversely, patients with lesions to the RDL/M region showed
significant facilitation of both concordant and discordant meanings
relative to baseline. Unlike controls and patients with damage restricted
to the RM region, these patients primed to both conditions, regardless of
the preceding context. This may indicate that the right dorsolateral
frontal cortex, alone or in combination with right medial regions,
contributes to the selection of contextually appropriate meanings and that
selective access is abolished when the right dorsolateral lobe is damaged.
As a result of problems abolishing the contextually inappropriate meaning,
facilitation appears equivalent for both concordant and discordant
meanings. Impaired ability to abolish or filter irrelevant information
following right dorsolateral damage is consistent with other research that
has implicated this region in response selection (Braver et al., 2001; Perlstein et
al., 2003) and in filtering irrelevant information in selective
attention (Stuss et al., 2002). The results of
this study suggest that the involvement of the right dorsolateral region
in selective access to relevant information might also extend to
lexical-semantic priming. It is important to note, however, that when
patients with pure RM and pure right dorsolateral lesions were directly
compared to one another, there was not a statistically significant
difference in their priming effects in the discordant condition. Instead,
there was a trend for patients with pure right dorsolateral damage to show
greater facilitation in the discordant meaning relative to the pure RM
group. There are two possible explanations for this lack of a difference.
First, it may be that damage to the right dorsolateral region is
necessary, but not sufficient to cause a disruption in lexical-semantic
priming. Second, it is possible that our sample size was too small to
detect a true difference. Because our individual participant analyses
revealed that all three of the patients in the pure dorsolateral group
showed facilitation of the discordant meaning, whereas only 1/6 of the
RM patients demonstrated facilitation, we believe that the latter
explanation contributed to our lack of a group difference.
Unlike our RDL/M patients, those with RM damage did not show
facilitation of the contextually inappropriate meaning relative to
baseline. However, patients with RM damage in our study also failed to
show facilitation of the context-appropriate meaning of the homograph, and
actually showed a trend toward suppression of the appropriate
meaning, p = .08. The fact that they failed to show facilitation
of either word meaning is difficult to explain since they showed
significant facilitation in the neutral condition—a condition that
theoretically produces less priming than the concordant condition
in healthy controls (Copland et al., 2000a).
While these data suggest that patients with RM damage also show impaired
lexical-semantic priming, their pattern is very different from that
observed in patients with RDL/M lesions and may indicate a more basic
problem with semantic processing (i.e., difficulty accessing and/or
initiating a response to either meaning in the presence of contextual
constraints).
On the surface, the results from our study and the study by Metzler
(2001) suggest that controls and patients with
RM frontal lesions show the expected pattern of slower responses to the
context-inappropriate relative to the appropriate meaning. However, the
results of our study are not entirely consistent with those found by
Metzler. Specifically, when we reanalyzed our data using the same approach
as Metzler, we found that our RM group did not demonstrate the abnormal
“hyperpriming” effect that she reported, but rather, the RM
patients in the present study look much like the controls. In addition, by
including a baseline condition, we demonstrated that those with RM lesions
do not show normal facilitation for the contextually appropriate homograph
meaning. Therefore, whether or not the RM group shows an abnormal pattern
of semantic priming may depend on how the data are analyzed (i.e.,
inclusion of a baseline versus proportional scores). Another
possibility, however, is that our use of multiple ISIs versus
Metzler's use of a single ISI may have differentially impacted
patient performances. Finally, our divergent findings may reflect
differences between the patient groups in each study. In particular,
Metzler's patients were of diverse etiologies (i.e., multiple
sclerosis, closed head injury) and the lesion locations were uncertain in
several patients. In contrast, all of our patients had unilateral frontal
resections for treatment of intractable epilepsy. Therefore, it is
possible that methodological and/or patient differences account for
the discrepancies between our results and those of Metzler.
Despite differences in the exact nature of semantic-priming effects,
it is important to point out that both our study and that of
Metzler's provide support for the role of the right frontal lobe in
semantic priming, that is, lexical-ambiguity resolution. These data are
consistent with much of the divided visual field literature with healthy
controls implicating the right hemisphere, more generally, in aspects of
lexical-ambiguity resolution (Chiarello & Richards,
1992; Coney & Evans, 2000; Faust & Chiarello, 1998; Nakagawa, 1991; Titone,
1998) and with a handful of patient studies suggesting that the
right hemisphere is involved in suppressing contextually inappropriate
meanings of homographs in discourse comprehension (Tompkins et al., 1997, 2000, 2001). Furthermore, our data provide evidence that
regional differences in semantic priming exist within the right frontal
lobe, a finding that has not been reported in the patient literature
previously.
On the other hand, results from neuroimaging studies have been mixed
and implicate a variety of frontal and temporal regions in
lexical-semantic priming (for a review, see Henson,
2003). In the only functional magnetic resonance imaging (fMRI)
study that we found using lexically ambiguous primes, Copland et al.
(2003) reported decreased activations in the
left middle temporal and left inferior prefrontal regions thought to
reflect semantic priming, but no evidence of right prefrontal involvement.
In an auditory lexical-decision task, Kotz et al. (2002) reported decreased activations in several
frontal and temporal lobe regions, including the right and left deep
frontal operculum, for unrelated compared to related target words. Other
researchers have found decreased activations in the superior and middle
temporal gyri, left precentral gyrus, right caudate (Rissman et al., 2003) and left anterior temporal lobe
(Rossell et al., 2003), in studies of semantic
priming. Thus, results from neuroimaging studies have been quite mixed and
suggest that a network of regions within the left and right hemisphere may
contribute to lexical-semantic priming. These studies, however, have
generally employed tasks that were quite different from one other and from
the task used in the present study (i.e., lexical decision vs.
word naming; use of different ISIs and stimulus-onset asynchronies (SOAs);
auditory vs. visual word presentation; ambiguous vs.
nonambiguous primes). In addition, our use of a “recall”
condition within the experimental paradigm differs from previous studies
of lexical-semantic priming. Therefore, it is possible that methodological
differences contribute to the variability within the neuroimaging
literature and may explain inconsistencies between existing fMRI priming
studies and ours.
Another point that deserves mention is that, unlike many studies
(Copland et al., 2000a; Hagoort, 1993), we did not obtain an effect for ISI.
This finding contrasts with some literature on lexical-ambiguity
resolution that suggests that both meanings of an ambiguous word are
initially activated, followed by suppression, or decreased activation, of
the contextually inappropriate meaning. Conversely, our results are
consistent with other literature suggesting that healthy controls can use
context to select meaning across all stages of lexical-ambiguity
resolution (Coney & Evans, 2000; Grinrod & Baum, 2002; Simpson
& Kreuger, 1991) and this may be especially the case when the
context created is sufficiently constraining (Tabossi
et al., 1987). However, it is possible that the repetitive nature
of the stimuli in our experiment resulted in expectancy effects that
diluted the effect of our ISI manipulation. It is of note, however, that
since repetition effects did not appear to differ significantly among the
groups, expectancy effects cannot explain the most interesting findings in
our study, that is, group differences in priming. Other methodological
differences between our study and existing studies may also account for
these discrepancies with regard to ISI. In particular, our use of question
marks as opposed to nonwords and/or the use of a word reading as
opposed to a button press response may have affected our results. In
addition, we included a recall condition that may have invoked more
controlled processing at the early ISI and accounted, in part, for the
discrepancies between our study and previous studies.
Finally, although we feel that our results provide evidence for the
disruption of lexical-semantic priming in patients with right frontal
lesions, these conclusions should be viewed with caution given our small
sample sizes (particularly in the subgroup analyses). For example,
although we did not find a significant main effect or any
interactions with ISI, the main effect of ISI approached significance
(p = .061). Therefore, it is possible that our study was somewhat
underpowered and that the effect of ISI would have reached significance
with a larger patient sample. Similarly, there was a trend for the groups
to differ in their reaction times to the baseline condition from which we
computed our priming scores (p = .057). Significant differences
among the groups in the baseline condition somewhat complicates the
interpretation of between-group differences in priming. However, it is
important to point out that between-group differences in the baseline
condition cannot explain the different pattern of
lexical-semantic priming within each group.
In summary, our results suggest that healthy controls use context to
guide meaning selection when resolving ambiguities in language, and
therefore, they show normal lexical-semantic priming. Conversely, patients
with right frontal lobe damage show disrupted patterns of lexical-semantic
priming that differ depending on the lesion location within the right
frontal lobe. These results, however, are very preliminary and should be
replicated in a larger sample of patients with well-circumscribed, right
medial and dorsolateral frontal lobe damage.
ACKNOWLEDGMENTS
This project was supported by the Epilepsy Foundation through the
generous support of the American Epilepsy Society and by an American
Psychological Association Dissertation Award.
