Contextual modulation of N400 amplitude to lexically ambiguous words

Brain and Cognition 55 (2004) 470 478 www.elsevier.com/locate/b&c Contextual modulation of N400 amplitude to lexically ambiguous words Debra A. Titone a, * and Dean F. Salisbury b a Department of Psychology, McGill University, Stewart Biological Sciences Building, 1205 Dr. Penfield Ave., Montreal, Que., Canada H3A 1B1 b Department of Psychiatry, Harvard Medical School, McLean Hospital, USA Accepted 12 February 2004 Available online 12 April 2004 Abstract Though much is known about the N400 component, an event-related EEG potential that is sensitive to semantic manipulations, it is unclear whether modulations of N400 amplitude reflect automatic processing, controlled processing, or both. We examined this issue using a semantic judgment task that manipulated local and global contextual cues. Word triplets (prime noun target, e.g., finance bank money) were sequentially presented on a computer screen (500 ms duration, 1000 ms stimulus onset asynchrony), in which the second word was a homograph. The first word (prime) created a neutral-, dominant-meaning- or subordinate-meaningbiased global context, and the third word (target) created a dominant- or subordinate-biased local context that was either congruent or incongruent with the global context established by the first prime word. Participants were instructed to read all three words but to decide only whether the second and third words were semantically related. Event-related potentials (ERPs), specifically the N400, were recorded to the third terminal word. N400 amplitudes evoked by dominant meaning-related third words incongruent with the globally biased subordinate context (e.g., river bridge money) were significantly more negative than dominant endings in neutral contexts (e.g., taxi bank money), but not different from unrelated filler triplets. In addition, there was some indication that left hemisphere, temporal parietal electrode sites were associated with greater N400 negativity for dominant targets in conflicting subordinate global contexts than homologous right hemisphere electrode sites, the latter of which showed significant activation to subordinate meanings in cooperating contexts. Thus, N400 amplitude was more affected by global than local context suggesting that controlled processes may take priority over automatic processes in modulating N400 amplitude, especially for left hemisphere electrode sites. Ó 2004 Elsevier Inc. All rights reserved. 1. Introduction The N400 is a negative-going potential that occurs approximately 400 ms following a semantically based stimulus event (reviewed in Kutas & Federmeier, 2000). Its amplitude is inversely related to the degree of semantic relatedness of a stimulus event (e.g., The man liked his coffee with dog elicits a larger (i.e., more negative) N400 amplitude than The man liked his coffee with sugar). Consequently, unrelated word pairs, incongruous sentence endings, and congruous but unexpected sentence endings all elicit larger N400 responses than, respectively, related word pairs, congruous sentence endings, and * Corresponding author. Fax: +1-514-398-4896. E-mail address: dtitone@psych.mcgill.ca (D.A. Titone). congruous but expected sentence endings. Modulations of N400 amplitude have also been found for pictures and faces (e.g., Debruille, Pineda, & Renault, 1996; Nigam, Hoffman, & Simons, 1992), which are semantically meaningful, or at the very least, verbally classifiable. Though much is known about the many linguistic variables that modulate N400 amplitude, its functional significance from a cognitive and psycholinguistic standpoint remains a matter of debate. Given the higher-order linguistic variables that affect N400, some researchers claim that modulations of N400 amplitude reflect sentence integration exclusively (e.g., van den Brink, Brown, & Hagoort, 2001). According to this view, experimental manipulations that modulate N400 amplitude may be classed as instances of post-lexical or controlled semantic processing. For example, using a 0278-2626/$ - see front matter Ó 2004 Elsevier Inc. All rights reserved. doi:10.1016/j.bandc.2004.02.073

D.A. Titone, D.F. Salisbury / Brain and Cognition 55 (2004) 470 478 471 semantic priming paradigm, Holcomb (1993) found that changes in N400 amplitude were not coupled with changes in semantic priming RT effects for perceptually degraded targets, which would have been predicted by automatic accounts of N400, given that stimulus degradation is thought to affect lexical selection. Thus, N400 amplitude was argued to reflect contextual integration after lexical selection had taken place. In contrast, other researchers suggest that automatic semantic processing, in addition to controlled semantic processing, modulates N400 amplitude. For example, Deacon, Hewitt, Yang, and Nagata (2000) found smaller N400 amplitudes to semantically related word pairs when the first word was not consciously perceived, and presumably, not subject to controlled semantic processing. In an earlier study, Deacon, Uhm, Ritter, Hewitt, and Dynowska (1999) found that automatic and expectancy-based priming effects on N400 amplitude corresponded to the pattern of behavioral results from the classic Neely (1977) experiment. As part of this experiment, participants are instructed to expect members of one category (e.g., types of furniture) when they encounter members of another category (e.g., body parts). Thus, automatic and expectancy-based cues can be put into conflict on low frequency switch trials where an unexpected semantic associate follows the prime (e.g., arm leg). Automatic priming effects alone were obtained at a short stimulus onset asynchrony (SOA 250 ms), whereas both automatic and expectancy-based priming effects on N400 amplitude were obtained at a longer SOA (2000 ms). Thus, a conservative view of the N400, consistent with the conclusions of Kutas and Federmeier (2000), is that automatic and controlled semantic processes jointly affect N400 amplitude, although they may follow different time courses or be differentially weighted as a function of the specific comprehension problem to be solved. It may be argued that the automatic and expectancybased cues employed in Deacon et al. (1999) and in the original Neely (1977) experiment can be recast as local and global contextual cues, respectively. Thus, the local context corresponds to the actual word that is encountered, whereas the global context corresponds to the task instruction whereby participants come to expect exemplars of one category to follow exemplars of another. Although these specific local and global contextual cues are somewhat artificial, language users routinely encounter more naturally occurring local and global contextual cues during comprehension that may affect N400 amplitude, and thus shed light on its functional significance. In particular, several studies indicate that global context reduces (i.e., makes more positive) the N400 potential that would otherwise be evoked by particular sentences. In a discourse processing experiment discussed by Kutas and Federmeier (2000), for example, van Berkum, Hagoort, and Brown (1999) found a smaller N400 response to words in local sentence contexts as a function of the global discourse context in which a sentence occurred. In that study, the words quickly and slowly elicit identical N400 responses in the sentence frame The mouse quickly/slowly returned to its hole. However, quickly elicits a more positive-going N400 when the sentence is preceded by the sentence The cat entered the room suddenly startling a mouse which had found a bit of cheese in the corner, suggesting that it had been primed by the global context. Thus, global context appears to reduce (i.e., make more positive) N400 amplitude over and above that of a wordõs local contextual environment. In another experiment that examined healthy participants (and schizophrenia patients), Sitnikova, Salisbury, Kuperberg, and Holcomb (2002) found effects of local vs. global context on N400 amplitude for ambiguous words (e.g., bridge) in sentences for healthy controls (but interestingly, not for schizophrenia patients; see also Titone, Levy, & Holzman, 2000, for a similar pattern of behavioral data). Here, sentence constraint (the global context) was put into conflict with the inherent semantic relationship between two words in the sentence (the local context, e.g., The guest played bridge because the river had rocks in it.) The results showed that the N400 evoked by the word river was more negative when the sentence context biased bridgeõs subordinate meaning (i.e., card game) than when it biased its dominant meaning (i.e., structure spanning water) (see also, Van Petten & Kutas, 1987). However, because this study did not include a neutral baseline condition where the global context neither cooperated nor conflicted with the local word-level relationship, it is unclear whether contextual modulation of N400 amplitude was facilitory or inhibitory in nature. This issue is important because it addresses the potential roles of automatic and controlled processing in modulating N400 amplitude. The purpose of the present study is to further examine the simultaneous effects of local and global context on N400 amplitude to determine whether the N400 potential evoked by semantically related words can be enhanced or inhibited by a cooperating or conflicting global context. Similar to Sitnikova et al. (2002), the present study examined ambiguous words in dominant or subordinate biased contexts. Unlike Sitnikova et al. (2002), we examined the effects of cooperating and conflicting contexts for isolated words with respect to a neutral baseline condition to determine whether global context would increase or reduce N400 amplitude. Specifically, ambiguous homographs such as bank were embedded in word triplets where the first word was either neutral with respect to the homographsõ dominant and subordinate meanings, biased toward dominant meanings, or biased toward subordinate meanings. The homographs comprised the second word of each triplet,

472 D.A. Titone, D.F. Salisbury / Brain and Cognition 55 (2004) 470 478 and the final target word was related to either the dominant or subordinate meaning (e.g., finance bank money). Thus, we will examine facilitation of the N400 effect to a global cooperating context (i.e., a more positive component than that found for a neutral context), and/or inhibition to a global conflicting context (i.e., a more negative component than that found for a neutral context) for the same two semantically related words, which themselves comprise the local context. We assume that automatic processing will be manifested by an effect of local context (i.e., the relationship between the second and third words) that resists any global contextual influence, whereas controlled processing will be manifested by an effect of global context that directly affects processing of the semantic relationship between the second and third words, and the N400 response to those two words. Further, the pattern of data for conflicting global contexts (e.g., river bank money) is likely to be most important for evaluating the role of automatic and controlled processes in modulating N400 amplitude, given that automatic and controlled processing accounts do not differ substantially in their predictions for cooperating contexts (e.g., river bank muddy). That is, both accounts would be consistent with a facilitory effect of the cooperating context, or no difference between the cooperating and neutral contexts (assuming that a particular meaning was maximally activated in the neutral context). Thus, the absence of decreased N400 amplitude to the third word in a conflicting global context, and the presence of increased N400 amplitude to the same word relative to the unrelated condition, would suggest that the local semantic relationship between the final two words alone modulates N400 amplitude. This pattern of data (i.e., cooperating context P neutral context ¼ conflicting context > unrelated condition) would attribute priority to automatic processes in modulating N400 amplitude. In contrast, the presence of decreased N400 amplitude to the third word in a conflicting global context relative to a neutral context, but no difference in N400 amplitude to the same third word relative to an unrelated condition, would suggest that the global context provided by the first word modulates N400 amplitude over and above the automatic processing of the local context (i.e., the inherent semantic relationship between the final two words). This pattern (i.e., cooperating context P neutral context > conflicting context ¼ unrelated condition) would attribute priority to controlled processes in modulating N400 amplitude. Finally, decreased N400 amplitude to the third word in a conflicting global context relative to a neutral context, but increased N400 amplitude to the same third word relative to the unrelated condition, would suggest that both the local semantic relationship between the final two words and the global context provided by the first word modulates N400 amplitude. This pattern (i.e., cooperating context P neutral context > conflicting context > unrelated condition) would support the view that both automatic and controlled processes modulate N400 amplitude. 2. Methods 2.1. Participants Twenty-four right-handed men and women were recruited through newspaper advertisements from the local community. All participants were native English speakers. None of the participants had a history of drug dependence, neurological disease or trauma, dementia, delirium, psychopathology, and any immediate family history of psychopathology (by self-report). All subjects gave their informed consent. 2.2. Stimuli Word triplets requiring a yes response consisted of 126 homographs preceded by neutral, dominant- or subordinate-biased context words, and followed by dominant- or subordinate-biased words (respectively, taxi bank money, finance bank money, river bank money, taxi bank muddy, finance bank muddy, river bank muddy). All homographs had clear dominant and subordinate meanings as determined by a greater than 1.5 ratio of dominant to subordinate responses from a large corpus of free association norms (Nelson, McEvoy, Walling, & Wheeler, 1980). Semantically related context and target words were selected also using these norms. Thus, two factors were crossed in constructing each yes triplet; target type (i.e., the third word which was either dominant, subordinate) and context type (i.e., the first word which was either cooperating, neutral, conflicting with respect to the semantic relationship emphasized by the final two words). The 126 homographs were counterbalanced across six stimulus lists so that by the end of the experiment, every homograph had an equal chance of participating in each condition, and all participants received an equal number of items per condition ðn ¼ 21Þ, without encountering repetitions of any one homograph. Word triplets requiring a no response were also presented, and these triplets were matched in their relational structure to the yes triplets with the exception of the final two words on which the semantic judgment decision was based. Although we collapsed across all no triplets in the final analysis, it was important to create these different categories of no triplets so that participants would not develop any response strategies (e.g., say yes whenever the first two words are related). Consider first the relational structure for the

D.A. Titone, D.F. Salisbury / Brain and Cognition 55 (2004) 470 478 473 cooperating context triplets (e.g., river bank muddy). Here, the first word is related to the second and third words, and the second word is related to the third word. Thus, in the corresponding no item, we created triplets where all the same relationships existed except the relationship between the second and third word (e.g., shed hair roof i.e., shed is a homograph that is related to both hair and roof, but hair and roof are not themselves related). Consider next the relational structure for the conflicting context triplets (e.g., river bank money). Here, the first word is related to the second word, and the second word is related to the third word, but the first and third words are unrelated. Thus, in the corresponding no item, we created triplets where the same relationship existed between the first and second words, but where the relationship between the second and third words was eliminated (e.g., salt pepper book). Finally, consider the relational structure for the neutral triplets in which only the final two words of each triplet were related (e.g., taxi bank money). Thus, in the corresponding no item, none of the three words were semantically related (e.g., dog table radio). There were 21 distinct instances of each no triplet condition. 2.3. Experimental procedure and analytic technique Words from each triplet were presented sequentially on a computer screen. Each trial sequence was as follows: first word duration 500 ms, 500 ms delay, second word duration 500 ms, 500 ms delay, third word duration 2500 ms (during which time a response occurred). The inter-trial interval was 500 ms. Participants were instructed to indicate as quickly and accurately as possible whether the final two words of the triplet were related by making a right thumb press for a yes response or a left thumb press for a no response. Participants sat approximately 1 m from the computer screen. Individual letters comprising the words were presented in lowercase font and 5 cm high. The entire list of words was between 6 and 12 cm wide. EEG activity was recorded from the scalp through 28 tin electrodes arranged in pre-configured caps (Electro- Cap International). Linked-earlobe electrodes served as the reference, and a forehead electrode served as the ground. Bipolar electrodes positioned medial to the right eye, one above and one below, were used to monitor vertical eye movements and blinks. Bipolar electrodes placed at the outer canthi of the eyes were used to monitor horizontal eye movements. All electrode impedances were below 3 kx, and the ears were matched within 1 kx. The EEG amplifier bandpass was 0.15 (6 db/octave rolloff) to 40 Hz (36 db/octave rolloff). Single trial epochs were digitized at 3.9 ms/sample (256 Hz). Each epoch was of 1100 ms duration, including a 100 ms pre-stimulus baseline. Averaging and artifact rejection were performed off-line. ERP responses were digitally low-pass filtered at 8.5 Hz with a 24 db/ octave rolloff to remove ambient electrical noise, muscle artifact, and alpha contamination. Epochs from each electrode site were baseline corrected by subtraction of the average pre-stimulus voltage, and corrected for eye movement artifact using regression-based weighting coefficients. Trials were baseline corrected after eyecorrection. Epochs that contained voltage levels exceeding 75 lv at F7, F8, Fp1, or Fp2 were rejected. Averages were computed for the final word of each yes triplet and for the final word of all unrelated noõ triplets. N400 amplitude was measured over a 50 ms window centered about 400 ms post-stimulus onset (i.e., 375 425 ms), which reflected the peak N400 latency for all conditions at Cz. Since electric fields arise simultaneously across the scalp, using the static interval centered about N400 peak latency assures that the same underlying cortical phenomena is measured at each site. Repeated measures ANOVAs on amplitudes of the N400 were computed separately for the midline electrodes (Fz, Cz, and Pz) and temporal parietal electrode sites (TCP1 and TCP2, corresponding to WernickeÕs left and right), with target type (dominant, subordinate; i.e., the final word) and context type (cooperating, neutral, conflicting, unrelated; i.e., the first word) as within subject factors. 2.4. Results All behavioral and N400 amplitude data are presented in Table 1. Grand average waveforms for the tripletsõ final words are presented in Figs. 1A and B. Note that three subjectsõ data were discarded due to equipment failure. 2.5. Semantic judgment responses Semantic judgment accuracy and response latencies of correct responses were analyzed using two separate 2 (Target type: dominant, subordinate) 4 (Context type: cooperating, neutral, conflicting, unrelated) repeated measure ANOVAs. Accuracy of semantic judgment responses varied as a function of both target type and context. Cooperating contexts failed to significantly improve semantic judgment accuracy for dominant targets (F ð1; 20Þ ¼ 3:1, p ¼ :08) and subordinate targets (F ð1; 20Þ ¼ 2:4, p ¼ :13). However, conflicting contexts significantly impaired semantic judgment accuracy for dominant targets (F ð1; 20Þ ¼7:0, p <:01) and subordinate targets (F ð1; 20Þ ¼25:6, p <:01) relative to the neutral context condition. Further, a significant interaction between target type and context indicated that conflicting contexts were more detrimental for subordinate targets than for dominant targets (F ð3; 60Þ ¼9:1, p <:01).

474 D.A. Titone, D.F. Salisbury / Brain and Cognition 55 (2004) 470 478 Table 1 Summary of behavioral and electrophysiological data DOM targets SUB targets UNREL targets Mean SEM Mean SEM Mean SEM Semantic judgment responses to final two words Latency (ms) 1237 57 Cooperating context 964 46 1048 46 Neutral context 963 41 1105 45 Conflicting context 1011 42 1190 52 Facilitation (Neut-Coop) )1 57 Inhibition (Neut-Confl) )48 )85 Percent correct 91.84 2.42 Cooperating context 93.42 1.93 83.22 2.32 Neutral context 89.80 2.06 80.05 2.56 Conflicting context 84.35 2.21 69.62 2.91 Facilitation (Neut-Coop) )3.63 )3.18 Inhibition (Neut-Confl) 5.44 10.43 N400 amplitude (microvolts) to final word Midline electrodes (Fz, Cz, Pz) )1.21 0.41 Cooperating context 1.78 0.61 0.11 0.60 Neutral context 1.31 0.49 )1.08 0.46 Conflicting context )0.41 0.49 )0.43 0.67 Facilitation (Neut-Coop) )0.47 )1.19 Inhibition (Neut-Confl) 1.72 )0.65 TCP1 (LH) 0.05 0.63 Cooperating context 1.74 0.90 0.90 0.74 Neutral context 1.17 0.75 0.19 0.63 Conflicting context )0.32 0.72 0.23 0.76 Facilitation (Neut-Coop) )0.57 )0.71 Inhibition (Neut-Confl) 1.49 )0.04 TCP2 (RH) Cooperating context 2.16 0.69 1.20 0.65 Neutral context 1.72 0.55 0.32 0.46 Conflicting context 0.65 0.75 0.26 0.77 )0.22 0.42 Facilitation (Neut-Coop) )0.44 )0.88 Inhibition (Neut-Confl) 1.08 0.06 * p <:05. ** p <:01. Latency of correct semantic judgment responses also varied as a function of target type and context. Cooperating contexts failed to significantly speed correct judgment responses for dominant targets relative to the neutral context condition (F ð1; 20Þ < 1). However, cooperating contexts did significantly speed responses for subordinate targets (F ð1; 20Þ ¼8:9, p <:01). Conflicting contexts slowed correct responses relative to the neutral context condition for both dominant and subordinate targets (respectively, F ð1; 20Þ ¼6:5, p <:05; F ð1; 20Þ ¼20:3, p <:01). As in the accuracy data, there was a significant interaction between target type and context indicating that subordinate targets were differentially more affected by conflicting contexts than dominant targets relative to the neutral context condition (F ð3; 60Þ ¼17:1, p <:01). 2.6. N400 amplitude 2.6.1. Midline electrode sites (Fz, Cz, Pz) N400 amplitudes to correct responses were analyzed using a 2 (Target type: dominant, subordinate) 4 (Context type: cooperating, neutral, conflicting, unrelated) 3 (Electrode site: Fz, Cz, Pz) repeated measures ANOVA. N400 amplitude varied as a function of all three independent variables. A main effect of target type indicated that dominant targets had significantly more positive N400 amplitude than subordinate targets (F ð1; 20Þ ¼6:7, p <:05). A main effect of electrode site indicated that N400 amplitude was more positive-going for all stimuli at posterior electrode sites than at frontal electrode sites (Fz < Cz < Pz) (F ð2; 40Þ ¼9:3, p <:01). A

D.A. Titone, D.F. Salisbury / Brain and Cognition 55 (2004) 470 478 475 Fig. 1. (A) Waveforms for event-related potentials evoked by dominant-related targets as a function of each context type (i.e., cooperating, neutral, and conflicting) and the final word of unrelated control triplets at different electrode sites. (B) Waveforms for event-related potentials evoked by subordinate-related targets as a function of each context type (i.e., cooperating, neutral, and conflicting) and the final word of unrelated control triplets at different electrode sites. main effect of context type indicated that the global context created by the first word of each triplet influenced the N400 amplitude evoked by the third word. However, this main effect was qualified by a marginal interaction between target type and context (F ð3; 60Þ ¼2:5, p ¼ :07). Cooperating context relative to the neutral context condition did not alter N400 amplitude for dominant targets (F ð1; 20Þ < 1) or subordinate targets (F ð1; 20Þ ¼ 2:4, p ¼ :12). In contrast, conflicting context relative to the neutral context condition were associated with more negative-going N400 amplitude for dominant targets (F ð1; 20Þ ¼5:1, p <:05) but not for subordinate targets (F ð1; 20Þ < 1) (i.e., the first word inducing the dominant meaning when the target is related to the subordinate

476 D.A. Titone, D.F. Salisbury / Brain and Cognition 55 (2004) 470 478 meaning). This differential effect of conflicting contexts for the two target types was also evident in a marginal interaction between target type and context (F ð3; 60Þ ¼ 2:5, p ¼ :07). Conflicting contexts did not differ from the unrelated condition for either target type (dominant and subordinate targets, respectively, F ð1; 20Þ ¼1:1, p ¼ :30; F ð1; 20Þ ¼1:0, p ¼ :31). 2.6.2. Temporal parietal electrode sites (TCP1, TCP2) N400 amplitude to correct responses was analyzed using a 2 (Target type: dominant, subordinate) 4 (Context type: cooperating, neutral, conflicting, unrelated) 2 (Electrode site: TCP1, TCP2) repeated measures ANOVA. In this analysis, we were particularly interested in whether hemispheric differences in electrode site emerged as a function of context. The only significant effect in the omnibus ANOVA was a main effect of context (F ð3; 60Þ ¼3:4, p <:05) indicating that N400 amplitude was more negative-going for the conflicting context and unrelated condition than for the cooperating context condition. However, planned comparisons between the context and target conditions as a function of electrode site (i.e., left vs. right hemisphere) revealed that contextual facilitation and inhibition differed as a function target type. Cooperating contexts for dominant targets did not affect N400 responses relative to the neutral context condition at either electrode site (TCP1, F ð1; 20Þ ¼1:6, p ¼ :22; TCP2, F ð1; 20Þ < 1). In contrast, cooperating contexts for subordinate targets were associated with more positive-going N400 amplitude relative to the neutral context condition, especially at the right hemisphere electrode site (TCP1, F ð1; 20Þ ¼ 3:0, p ¼ :09; TCP2, F ð1; 20Þ ¼4:6, p <:05) In contrast with cooperating contexts, conflicting contexts for dominant meanings were associated with more negative-going N400 amplitude relative to the neutral context condition for both the left hemisphere and right hemisphere electrode site (TCP1, F ð1; 20Þ ¼10:8, p <:01; TCP2, F ð1; 20Þ ¼5:7, p <:05). Interestingly, a direct comparison of the conflicting context conditions for left and right hemisphere electrode sites revealed more negative-going N400 amplitude for the left hemisphere than for the right hemisphere (F ð1; 20Þ ¼4:5, p <:05). Conflicting contexts failed to affect N400 amplitude relative to the neutral condition for subordinate targets ðf < 1Þ. Note that these hemispheric effects were only found in the planned comparisons, thus their interpretation should be viewed with caution. 3. Discussion The purpose of the study was to examine whether N400 amplitude is jointly modulated by global and local contextual cues in an overt semantic judgment task where participants were instructed to attend to local contextual cues. The results confirmed these predictions in revealing a complex interaction of local and global contextual effects on N400 amplitude. First, modulations of N400 amplitude were only found for triplets containing dominant targets. None of the triplets containing subordinate targets produced N400 amplitudes that were more positive than the unrelated triplet condition, with the one exception of subordinate targets in cooperating contexts at right hemisphere, temporal parietal electrode sites. This suggests that dominant meanings have a clear priority over subordinate meanings with respect to N400 modulation, even for contexts that globally bias subordinate meanings (see also Salisbury, Shenton, Nestor, & McCarley, 2002). The association of right hemisphere temporal parietal electrode sites with more positive-going N400 amplitudes for cooperating contexts is consistent with other work suggesting a specific right hemisphere involvement in generating and maintaining activation of peripheral word meanings (e.g., Atchley, Burgess, & Keeney, 1999; Beeman, 1993; Brownell, 1988; Chiarello, 1991; Federmeier & Kutas, 1999; Titone, 1998), although given that this effect was only found in planned comparisons, further work will be necessary to substantiate this interpretation of the data. Second, with respect to triplets containing dominant targets, only neutral and cooperating contexts produced significantly smaller N400 amplitudes than the unrelated condition. Specifically, conflicting context triplets containing dominant targets (e.g., river bank money) evoked more negative-going N400 amplitudes than neutral context triplets sharing the same local dominant context (e.g., taxi bank money.) Moreover, N400 amplitude for these conflicting triplets did not differ from that found for no triplets where the second and third words were unrelated. Further, N400 amplitude for conflicting triplets appeared to be more negative at left hemisphere temporal parietal electrode sites (TCP1) than at right hemisphere sites (TCP2), consistent with previous work suggesting a specific left hemisphere involvement in lexical selection or inhibition, and semantic integration (e.g., Atchley et al., 1999; Beeman, 1993; Brownell, 1988; Chiarello, 1991; Federmeier & Kutas, 1999; Titone, 1998). Note that this hemispheric effect is also preliminary in that it was only statistically substantiated by planned comparisons and not the overall ANOVA. In contrast to the conflicting context triplets, the cooperating context triplets (e.g., finance bank money) did not differ in N400 amplitude from the neutral context triplets (e.g., taxi bank money). Taken together, these data suggest that global context trumps N400 priming normally evoked by a local context consisting of two semantically related words. These data also suggest that the dominant meaning was maximally activated in the neutral context condition given that

D.A. Titone, D.F. Salisbury / Brain and Cognition 55 (2004) 470 478 477 there was no increase in N400 amplitude in the cooperating context condition. The N400 results for dominant targets are consistent with the semantic judgment accuracy and latency data, although the behavioral results for subordinate targets show clear effects of context that are not found in the N400 data. In particular, conflicting contexts were associated with reduced accuracy and longer latencies of semantic judgment responses compared to the neutral context for triplets containing both dominant and subordinate targets. Cooperating contexts, on the other hand, produced a different pattern of results for dominant and subordinate targets. For dominant targets, cooperating contexts were associated with increased accuracy but no difference in latency compared to neutral contexts. For subordinate targets, cooperating contexts were associated with shorter latencies but no difference in accuracy compared to neutral contexts. In addition, semantic judgment responses to dominant targets were more accurate and faster than responses to subordinate targets, consistent with the behavioral literature on lexical ambiguity resolution (for comprehensive reviews, see Klepousniotou, 2002; Titone, 1998; Van Petten, 2002). This is exemplified by the finding of comparable accuracy for dominant meanings in conflicting contexts and subordinate meanings in cooperating contexts (see Table 1). N400 amplitude was also significantly smaller for dominant targets in neutral contexts than for subordinate targets in neutral contexts. The overall pattern of results is consistent with the view that N400 amplitude is more affected by global than by local semantic context, lending support to the notion that controlled semantic processes exert a greater influence on N400 amplitude than automatic semantic processes. We arrive at this conclusion for three main reasons. First, N400 amplitude to dominant targets in conflicting contexts was significantly larger (i.e., more negative) than N400 amplitude to dominant targets in neutral contexts. Second, N400 amplitude to dominant targets in conflicting contexts was no different from N400 amplitude of the unrelated no triplets. Finally, the task was inherently biased to produce an effect of local context because participants were instructed to focus their attention on the local semantic relationship between the tripletsõ final two words. If automatic semantic processes partially drove N400 amplitude, the conflicting context triplets would have shown smaller (i.e., more positive) N400 amplitudes relative to the unrelated condition, especially given that these triplets contained two semantically related words upon which explicit attention was directed for a semantic judgment response. It is possible that modifications of the experimental paradigm would evoke smaller N400 amplitudes for dominant targets in conflicting contexts than the unrelated condition, and thus an N400 effect for local contexts. First, the use of shorter SOAs than what was employed here (1000 ms) might be more conducive to observing an effect of local context, and thus, automatic processing on N400 amplitudes. However, we are confident that the 1000 ms SOA used here could have detected an effect of automatic processing on N400 amplitude given that Deacon et al. (1999) found automatic processing effects in a slightly different paradigm using an SOA of 2000 ms. Second, the semantic judgment task may have caused participants to encode the context word of each triplet more deeply than what would normally occur during passive reading, thus inducing an overall controlled processing strategy. Although given that participants were explicitly instructed to attend to the local context (i.e., the final two words), this feature of the task might also account for the data that were observed. Third, the use of single words rather than sentence contexts may increase the likelihood that these effects are occurring intra-lexically rather than semantically (e.g., Seidenberg, Tanenhaus, Leiman, & Bienkowski, 1982). For these reasons, future work will need to determine which of these and other potential boundary conditions are important factors in the modulation of N400 amplitude, or perhaps different subcomponents of the N400 potential (Kutas & Federmeier, 2000). Acknowledgments This study was supported in part by grants from NSERC, CRC, CFI, NIMH (R03 MH60272), and a NARSAD Young Investigator Award to D. Titone, and grants NIMH (RO158704), and a NARSAD Young Investigator Award to D. Salisbury. The authors are grateful to the participants, and also to Tali Ditman, Nicola Sumorok, and Margaret Fagan for assistance with participant testing and data analysis. References Atchley, R. A., Burgess, C., & Keeney, M. (1999). The effect of time course and context on the facilitation of semantic features in the cerebral hemispheres. Neuropsychology, 13, 389 403. Beeman, M. (1993). Semantic processing in the right hemisphere may contribute to drawing inferences from discourse. Brain and Language, 44, 80 120. Brownell, H. H. (1988). Appreciation of metaphoric and connotative word meaning in brain damaged patients. In C. Chiarello (Ed.), Right hemisphere contributions to lexical semantics. New York: Springer-Verlag. Chiarello, C. (1991). Interpretation of word meanings by the cerebral hemispheres: One is not enough. In P. Schwanenflugel (Ed.), The psychology of word meanings. LEA Publishers. Deacon, D., Hewitt, S., Yang, C., & Nagata, M. (2000). Event-related potential indices of semantic priming using masked and unmasked words: Evidence that the N400 does not reflect a post-lexical process. Cognitive Brain Research, 9, 137 146.

478 D.A. Titone, D.F. Salisbury / Brain and Cognition 55 (2004) 470 478 Deacon, D., Uhm, T. J., Ritter, W., Hewitt, S., & Dynowska, A. (1999). The lifetime of automatic semantic priming effects may exceed two seconds. Cognitive Brain Research, 7, 465 472. Debruille, J. B., Pineda, J., & Renault, B. (1996). N400-like potentials elicited by faces and knowledge inhibition. Cognitive Brain Research, 4, 133 144. Federmeier, K. D., & Kutas, M. (1999). Right words and left words: Electrophysiological evidence for hemispheric differences in meaning processing. Cognitive Brain Research, 8, 373 392. Holcomb, P. (1993). Semantic priming and stimulus degradation: Implications for the role of the N400 in language processing. Psychophysiology, 30, 785 806. Klepousniotou, E. (2002). The processing of lexical ambiguity: Homonymy and polysemy in the mental lexicon. Brain and Language, 81, 205 223. Kutas, M., & Federmeier, K. D. (2000). Electrophysiology reveals semantic memory use in language comprehension. Trends in Cognitive Science, 4, 463 470. Neely, J. H. (1977). Semantic priming and retrieval from lexical memory: Roles of inhibitionless spreading activation and limitedcapacity attention. Journal of Experimental Psychology: General, 106, 226 254. Nelson, D. L., McEvoy, C. L., Walling, J. R., & Wheeler, J. W. (1980). The University of South Florida homograph norms. Behavior Research Methods and Instrumentation, 12, 16 37. Nigam, A., Hoffman, J. E., & Simons, R. F. (1992). N400 to semantically anomalous pictures and words. Journal of Cognitive Neuroscience, 4, 15 22. Salisbury, D., Shenton, M. E., Nestor, P. G., & McCarley, R. W. (2002). Semantic bias, homograph comprehension, and eventrelated potentials in schizophrenia. Clinical Neurophysiology, 113, 383 395. Seidenberg, M. S., Tanenhaus, M. K., Leiman, J. M., & Bienkowski, M. (1982). Automatic access of the meanings of ambiguous words in context: Some limitations of knowledge based processing. Cognitive Psychology, 14, 489 537. Sitnikova, T., Salisbury, D. F., Kuperberg, G., & Holcomb, P. I. (2002). Electrophysiological insights into language processing in schizophrenia. Psychophysiology, 39, 851 860. Titone, D. (1998). Hemispheric differences in context sensitivity during lexical ambiguity resolution. Brain and Language, 65, 361 394. Titone, D., Levy, D., & Holzman, P. S. (2000). Contextual insensitivity in schizophrenic language processing: Evidence from lexical ambiguity. Journal of Abnormal Psychology, 109, 761 767. van Berkum, J. J., Hagoort, P., & Brown, C. M. (1999). Semantic integration in sentences and discourse: Evidence from the N400. Journal of Cognitive Neuroscience, 11, 657 671. van den Brink, D., Brown, C. M., & Hagoort, P. (2001). Electrophysiological evidence for early contextual influences during spokenword recognition: N200 versus N400 effects. Journal of Cognitive Neuroscience, 13, 967 985. Van Petten, C. (2002). Lexical ambiguity resolution. In L. Nadel (Ed.), Encyclopedia of cognitive science (pp. 867 872). London: Macmillan. Van Petten, C., & Kutas, M. (1987). Ambiguous words in context: An event-related potential analysis of the time course of meaning activation. Journal of Memory and Language, 26, 188 208.