- Email: [email protected]
N400 and category exemplar associative strength
International Journal of Psychophysiology 56 (2005) 45 – 54 www.elsevier.com/locate/ijpsycho
N400 and category exemplar associative strengthB M. Isabel Nu´n˜ez-Pen˜a*, M. Luisa Honrubia-Serrano University of Barcelona, Department of Behavioral Sciences Methods, Faculty of Psychology, Passeig Vall d’Hebron, 171, 08035 Barcelona, Spain Received 3 August 2004; received in revised form 14 September 2004; accepted 28 September 2004
Abstract The main purpose of this study was to examine if the N400 amplitude can be considered an index of category exemplar strength of association in a semantic categorization task. Series of six words from a particular semantic category were used as context, and the strength of association between the seventh word and the category was manipulated. Moreover, subjects were asked to perform two different tasks: one consisting of reading the words for comprehension and the other of making a decision regarding the congruence of the ending word. Results showed that the N400 component was elicited by whichever word that was not the best exemplar for the category (atypical members and nonmember of the category). No interaction between type of ending word and type of task was found around 400 ms poststimuli although the P3b component was present for infrequent stimuli in the decision task. It is concluded that the N400 amplitude is sensitive to category membership although no specific attention to the semantic relationship between words was required. D 2004 Elsevier B.V. All rights reserved. Keywords: N400; P3b; Semantic categories; ERPs
1. Introduction The interest in the nature of the processes underlying the N400 component has grown significantly since it was first described by Kutas and Hillyard (1980). These authors contrasted semantically conB Support: This research was supported by grants FI-PG/951.191 from the Generalitat de Catalunya, and BS02003-02440 from the Spanish Ministry of Science and Technology. * Corresponding author. Tel.: +34 93 312 58 53; fax: +34 93 402 13 59. E-mail address: [email protected] (M.I. Nu´n˜ez-Pen˜a).
0167-8760/$ - see front matter D 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.ijpsycho.2004.09.006
gruent ending words with semantically incongruent ending words in sentences and found that incongruent final words elicited a negative wave. This negativity reached its largest amplitude around 400 ms and showed a posterior distribution. This newly discovered component was called N400. Since then, many studies have been conducted to find out more about the nature of the N400 component (for a review, see Brown and Hagoort, 2000; Kutas and Federmeier, 2000; Kutas and Van Petten, 1994). Although N400 was first described as ban electrophysiological sign of the dreprocessingT of semantically anomalous informationQ (Kutas and
46
M.I. Nu´n˜ez-Pen˜a, M.L. Honrubia-Serrano / International Journal of Psychophysiology 56 (2005) 45–54
Hillyard, 1980, p. 203), it was later considered an bindex of the degree of semantic priming or activation that a word receives from the prior contextQ (Kutas et al., 1984, p. 216). In fact, Kutas and Hillyard (1984) found that the N400 amplitude was inversely related to the cloze probability of a word in a sentence (the cloze probability of a word is the frequency in which this word is produced as a completion of a particular sentence). The easier to integrate the word in the sentence, the smaller the N400 amplitude. This result showed that N400 was not merely an index of semantic incongruity because it was also elicited by semantically correct but less preferred endings. The N400 component is also sensitive to category membership. Several studies have shown that it is elicited whenever the presented word does not fit in a previously established semantic category. To date, N400 sensitivity to category membership has been observed both in sentences and in lists or pairs of words. Polich (1985) presented series of words belonging to the same semantic category and reported a negative wave whenever the ending word was not an exemplar of this category—in this case, the semantic category was established by the presentation of a series of words. Category effects on the N400 have also been observed during the performance of sentence verification tasks. In sentences like bA robin is a birdQ (Fischler et al., 1983) or bAll collies are dogs,Q (Kounios and Holcomb, 1992) the N400 component was elicited whenever the exemplar did not fit the category. In spite of the fact that true and false sentences were presented, the N400 component was independent of the truth–value of the sentence being modulated instead by the relatedness of the subject and predicate terms. The N400 component seems to be sensitive not only to category membership but also to the category exemplar associative strength. Heinze et al. (1998) presented subjects with pairs of words, the first of which was the name of a semantic category and the second a typical member, an atypical member or a nonmember of the category. These authors found that the N400 component was elicited for both atypical and nonmember words although the N400 amplitude was larger for the nonmembers. Similar results were obtained by Fujihara et al. (1998). Taken together, all
these results suggest that the semantic typicality of an exemplar modulates the N400 amplitude. Semantic categorization effects on the N400 have also been observed in the context of sentences. Federmeier and Kutas (1999) presented sentences that ended with the expected item, an unexpected item from the same category as the expected exemplar (within-category violations) or an unexpected item from a different category than the expected exemplar (between-category violations). The expected exemplar elicited a late positivity, but both violations elicited a different N400 response: a larger N400 amplitude for between- than for within-category violations (a similar result had been reported by Kutas et al., 1984). Federmeier and Kutas claimed that bsemantic features of the category exemplar (not necessarily a specific lexical item) most likely to complete the target sentence are activated prior to the presentation of the actual sentence-final word. When the prediction is incorrect and the expectancy is not met, the data are characterized by increased N400 activity relative to when the prediction is correctQ (Federmeier and Kutas, 1999, p.487). This explanation also accounts for the results by Heinze et al. (1998). Summarizing, the N400 seems to be a useful component to study semantic categorization effects because its amplitude is modulated by the degree of semantic relatedness between words. The aim of the present experiment was to study if the N400 amplitude is influenced by the degree of category exemplar associative strength. To this end, series of words belonging to the same category were used as context; the ending word could be a typical member, an atypical member or a nonmember of the category previously described. To our knowledge, this is the first study of any kind to examine the degree of category exemplar associative strength by using series of words to establish the semantic category. Cloze probability experiments have shown that the N400 amplitude seems to be an index of degree of semantic integration of a word in the previous context of a sentence. On the other hand, Federmeier and Kutas (1999) claimed that the N400 amplitude could be related to the activation of the semantic features of upcoming words prior to their occurrence. Our goal was to study if, given a series of words that define a semantic category, the amplitude of the N400 would be affected by the
M.I. Nu´n˜ez-Pen˜a, M.L. Honrubia-Serrano / International Journal of Psychophysiology 56 (2005) 45–54
category exemplar associative strength. If semantic features of the category are activated, then a N400 amplitude modulation is expected depending on the strength of association between the word and the category; the more incongruent the ending word presented with regard to the category, the larger the N400 amplitude. If a typical member of the category is activated, then it is expected to find an N400 of similar amplitude for atypical members and nonmembers of the category.
2. Materials and methods 2.1. Subjects Sixteen volunteers (four men; one left handed; mean age=24.8 years, standard error=1.5) participated in the experiment. They were native speakers of Spanish and had normal or corrected to normal vision. All gave informed consent to participate in the experiment. 2.2. Stimuli The stimuli consisted of 168 series of seven words. The first six words belonged to the same semantic category, and the seventh could be a typical member, an atypical member or a nonmember of the category previously established. These words were selected from the material produced by Soto et al. (1982). The procedure that Soto et al. followed for their material was similar to that used by Batting and Montague (1969). It consisted of asking the subjects to write down words belonging to the same semantic category in a notebook. Then total and partial frequencies were recorded for every word. Total frequency was computed by adding the number of times that a word was written in whatever position. Partial frequency, however, was computed by adding the number of times that a word was written in every position—first to seventh positions. The series was constructed as follows. In the first place, we selected words in second to seventh total frequency position for each category (compound words were never selected); these words were presented sequentially from lower to higher frequency and constituted the series. In the second place, the
47
ending word in the series was defined in the following way. The typical member was the word with highest total and partial frequency in a category. Usually, these two criteria matched; nevertheless, when this was not the case, the word with highest partial frequency was selected. To select the atypical member, the following two conditions had to occur: (1) the result of dividing the total frequency of the word selected as atypical member by the total frequency of the word that occupied the seventh position had to be around 0.2—words selected for this condition had a mean of 0.2, with a maximum of 0.3 and a minimum of 0.13; (2) the word had to present its higher partial frequency from third position. As for the nonmembers, these words were selected from the Alameda and Cuetos (1995) Spanish frequency dictionary following these criteria: (1) they were nouns; (2) they had high use frequency (mean=564.3; standard error=27.9); (3) they were nonmembers of the semantic category previously described; (4) they had not been selected as a part or ending of the series in the experiment. The experiment was controlled by the STIM 2.0 program (NeuroScan, Herndon, VA). Words were presented in uppercase and white color letters on a black background. The stimuli subtended a visual angle of 1.238 vertically and a mean of 4.98 horizontally. 2.3. Procedure Participants were tested in two experimental sessions carried out in separate days. The material presented in the two experimental sessions was the same, but participants were given different tasks in each. One task consisted of pressing one button on the mouse if the ending word was congruent with the previous series and the other button if the ending word was incongruent—the assignment of congruent and incongruent response to left or right button was counterbalanced across subjects (henceforth, the response task). In the other task, however, participants were instructed to read the words silently for comprehension because they had to answer a questionnaire about the stimuli at the end of the recording (henceforth, the reading task). The purpose of working with these two sessions was to control for possible decision effects on the ERPs. Each session lasted
48
M.I. Nu´n˜ez-Pen˜a, M.L. Honrubia-Serrano / International Journal of Psychophysiology 56 (2005) 45–54
about 2 h, and the order of sessions was counterbalanced across subjects. A minimum period of 30 days was left between sessions to minimize memory effects. Trials in both sessions consisted of series of seven sequentially presented words. A thousand ms after the ending word, an asterisk appeared on the screen giving the subject the following information: (1) the series had finished, (2) blinking was allowed, and (3) a response had to be emitted. Regarding this last point, motor response was only compulsory in the response task. Fig. 1 shows the outline of a trial. In both sessions, subjects sat in a comfortable armchair approximately 60 cm from a monitor. The session began with a short practice period to acclimatize them to the experimental conditions, consisting of several trials like those presented during the recording period. Participants were told to press one button if the ending word belonged to the previous semantic category and the other button if the ending word did not. They were then given feedback on the correctness of the response. Because our main aim was to ensure that subjects discriminated between experimental conditions, the training period did not finish until one of these learning criteria was met: (1) the participant had correctly answered the first 10 consecutive trials, or (2) the subject correctly answered 90% of trials. There was a 30-s rest after every 20 trials. When the training period was over, the recording period started. Subjects were instructed to relax and to keep their eyes on the screen. They were also told to avoid blinking; if they needed to blink, they should try to wait until an asterisk or a rest message appeared on the screen. Participants were presented with 126 trials in each session—42 of each type of stimulus— which were organized in 14 blocks of nine trials. Three trials of each type of stimulus were randomly presented in each block with the restriction that no more than three trials of the same type could be presented consecutively. Moreover, the same category was never presented twice inside a block. Participants
were given a 30-s rest after each block to allow them a time to blink. Words were presented for 500 ms with an interstimulus interval of 1000 ms. The intertrial period was 2000 ms. At the end of the recording in the reading task, participants were asked to answer both a memory and a recognition test. In the memory test, participants had to write down on a sheet of paper the name of the semantic categories that had been presented during the recording session; they were instructed to write as many categories as they could remember. In the recognition test, a 42-word list was presented, and subjects were requested to specify for each word if it had been previously presented during the recording session—32 of the 42 items in the word list had been previously presented during the recording session. The responses of these two tests allowed us to assess whether a participant had paid sufficient attention to the material. 2.4. EEG recording Recording and analysis of ERPs were performed with the SCAN 3.0 software (NeuroScan, Herndon, VA). EEG was recorded from seven tin electrodes mounted in a commercial electro-cap (Electro-Cap International, Eaton, OH) and positioned according to the 10–20 International System: C3, C4, P3, P4, Cz, Fz and Pz. Connected ear lobes served as reference, and an equidistant point between Fpz and Fz was used as location of the ground electrode. For monitoring eye movement and blinks, four additional electrodes were used: two for the VEOG recording placed above and below the right eye and two for the HEOG recording placed at the two external canthi. EEG and EOG channels were continuously digitized at a rate of 250 Hz by a SynAmpTM amplifier (5083 model, NeuroScan, Herndon, VA). A band pass filter was set from 0.05 to 30 Hz, and a notch filter of 50 Hz was used. Electrode impedance was always kept below 5 kV.
Fig. 1. Scheme of a nonmember ending trial.
M.I. Nu´n˜ez-Pen˜a, M.L. Honrubia-Serrano / International Journal of Psychophysiology 56 (2005) 45–54
2.5. Statistical analysis The statistical analysis was done in three steps: (1) analyses of the memory and the recognition test, (2) analyses of the behavioral data and (3) analyses of the electrophysiological data. Memory and recognition test were analyzed by means of descriptive statistics for the percentage of correctly remembered categories and the percentage of correctly recognized words. Behavioral data analyses could only be performed in the response task. Because participants had to give their motor response after the asterisk and not after the ending word of the series, reaction time was not a useful variable; therefore, number of errors was the only behavioral variable analyzed. Pressing the wrong button was considered an error; however, nonpressing and responses given outside the interval from 100 to 1900 ms were not counted as errors. Nonparametric statistics were used for the number of error analyses: first, the Friedman test was executed for the variable Type of ending word (typical member, atypical member and nonmember); second, the Wilcoxon test was performed for paired contrasts. The analysis of the electrophysiological responses was carried out according to this procedure. First, epochs were averaged for every subject in each experimental condition relative to a prestimulus baseline that comprised the 100 ms of activity preceding the epoch of interest. Trials with artifacts (voltage exceeded F50 Av in any channel) and those with response errors were excluded from the ERP averages. Second, ERP data were analyzed by computing the mean amplitude in 50 ms windows from 250 to 500 ms poststimuli and in 100 ms windows from 500 to 800 ms poststimuli. We analyzed 50 ms windows between 250–500 ms to identify early negativities and 100 ms windows between 500–800 ms to look for late positivities. The purpose of working with all these windows was to extract a detailed analysis of the waves. A 237 repeated measures analysis of variance (ANOVAs) was performed on the amplitude means, taking as independent variables Type of task (response or reading), Type of ending word (typical member, atypical member and nonmember) and Electrode (C3, C4, P3, P4, Fz, Cz and Pz). Repeated measures ANOVAs were performed with the Greenhouse–
49
Geisser correction (Geisser and Greenhouse, 1958) for sphericity departures, which was applied when appropriate; we reported the uncorrected degrees of freedom, the e value, the probability level following correction and the g 2 effect size index (Cohen, 1988; Hays, 1994; Kirk, 1996). Whenever a main effect reached statistical significance—p-value less than or equal to 0.05—, t-tests contrasts were calculated (Keselman and Keselman, 1988; Keselman and Keselman, 1993); to analyze statistically significant interaction, simple effects were calculated. In both cases, the Hochberg step-up approach was used to control for the increase in type I error (Hochberg, 1988; Keselman, 1998).
3. Results 3.1. Memory and recognition test Descriptive analyses of the memory and the recognition tests showed the following results. The mean of the percentage of correctly remembered categories was 62.8 (standard deviation=13.1; range 42.8–80.9). As for the percentage of correctly recognized words, the mean was 71.2 (standard deviation=10.3; range 57.1–88.1). These results allow us to conclude that participants understood the experimenter’s instructions and paid sufficient attention to the material during the reading task. 3.2. Behavioral data An ending word effect for the number of errors was obtained in the reading task (v 2[2, N=16]=10.43, p=0.005). Differences were found between typical and atypical members (Z= 2.99, p=0.003); there were more errors for atypical members. However, there were no statistically significant differences between nonmembers and both types of members. The median number of errors was 0, 4 and 1.5 for typical, atypical and nonmembers, respectively. 3.3. Electrophysiological data Grand average ERPs from all recording sites are shown in Fig. 2. In the response task, a negativity peak about 400 ms was present in atypical member
50
M.I. Nu´n˜ez-Pen˜a, M.L. Honrubia-Serrano / International Journal of Psychophysiology 56 (2005) 45–54
Fig. 2. Top: grand average ERPs for each type of ending word in the response task. Bottom: grand average ERPs for each type of ending word in the reading task.
and nonmember ending words. In the nonmember ending words, this negativity was followed by a positive wave peaking around 600 ms poststimulus and larger in centro-parietal sites. The pattern of voltage in the reading task was very similar to that described for the response task. However, a difference for the nonmember endings was present; in the reading task, the negative peak was not followed by the late positive component. In fact, no difference
seemed to exist in the reading task between the three types of ending word from 500 ms. 3.3.1. 200–500 ms interval A main effect of the Type of ending word was obtained in the 350–400 ms window ( F[2,30]=5.91, p=0.007, e=0.91, g 2=0.28) and the 400–450 ms window ( F[2,30]=8.35, p=0.001, e=0.98, g 2=0.35). Paired contrasts were conducted for both windows,
M.I. Nu´n˜ez-Pen˜a, M.L. Honrubia-Serrano / International Journal of Psychophysiology 56 (2005) 45–54 Table 1 Mean (in AV) and standard error (in brackets) for the different ending words presented 350–400 Typical member Atypical member Nonmember
400–450
1.08 (0.41) 0.13 (0.42) 0.61 (0.57)
0.49 (0.44) 0.84 (0.56) 1.41 (0.59)
Results are shown for the 350–400 and the 400–450 ms windows.
and amplitude differences were found between typical and atypical members and between typical members and nonmembers ( p-valuesb0.01 in each contrast); mean amplitude was more negative for atypical members and nonmembers than for members of the category at these windows. Mean amplitude for the different ending types is shown in Table 1. The interaction Series ending type Electrode reached statistical significance in the 350–400 ( F[12,180]=2.45, p=0.005, e=0.28, g 2=0.14) and the 400–450 ms window ( F[12,180]=2.92, p=0.001, e=0.30, g 2=0.16). Simple effect analyses showed that differences for the type of series ending were statistically significant in C4, Cz, P3, P4 and Pz. Therefore, the effect of the series ending type was only revealed in the centro-parietal sites. Paired contrasts were performed for the series ending type at each location, and the following results were found. In the 400–450 ms window, differences were found between typical and atypical members and between typical members and nonmembers ( p-valuesb0.01 in each contrast) at C4, Cz, P3, P4 and Pz. Again, amplitude was more negative for atypical members and nonmembers than for typical members of the category. In the 350–400 ms window, the same differences between typical members and nonmembers were found at C4, Cz, P3, P4 and Pz ( p-valuesb0.01 in each contrast). However,
51
differences between typical and atypical members were only present at P3 and Pz. Table 2 shows mean differences for the previous contrasts. The variable Type of task showed neither main effect nor interactions in the time windows in the 250– 500 ms interval. Therefore, it is concluded that Type of task had no effect on these time windows. 3.3.2. 500–800 ms interval The main effect of Type of ending word was reported in the 500–600 ms window ( F[2,30]=4.72, p=0.017, e=0.96, g 2=0.23), in the 600–700 ms window ( F[2,30]=11.88, pb0.001, e=0.98, g 2=0.42) and in the 700–800 ms window ( F[2,30]=11.79, pb0.001, e=0.88, g 2=0.44). Paired contrasts revealed that nonmember endings elicited a more positive voltage than the member endings in the interval between 500 and 800 ms poststimulus ( p-valuesb0.01). The Type of ending word effect was modulated by the Type of task—as the analysis of the Type of ending word Type of task interaction shows—in the 500–600 ms window ( F[2,30]=3.57, p=0.04, e=0.81, g 2=0.19), in the 600–700 ms window ( F[2,30]=6.87, p=0.003, e=0.78, g 2=0.31) and in the 700–800 ms window ( F[2,30]=4.27, p=0.023, e=0.91, g 2=0.22). Simple effect analyses showed that differences between reading and response tasks were present in nonmember endings but not in both types of member words. While there was no difference between response and reading task for member words, nonmember endings elicited a more positive voltage in the response task than in the reading task. Mean amplitude for each ending word in the three windows can be seen in Table 3. Differences between tasks of 2.88, 3.97 and 2.52 AV for the 500– 600, the 600–700 and the 700–800 ms windows, respectively, were reported when the ending word was a nonmember, whereas differences between tasks
Table 2 Mean differences (in AV) obtained after subtracting amplitudes for nonmembers and atypical members from amplitudes for typical members at every electrode Time (ms)
Contrast
C3
C4
P3
P4
Cz
Fz
Pz
350–400
Typical–atypical Typical–nonmember Typical–atypical Typical–nonmember
0.14 1.12 0.49 1.36
1.07 1.47* 1.41* 1.83*
1.35* 2.85* 1.91* 3.16*
1.01 1.81* 1.49* 2.26*
1.12 1.86* 1.38* 1.86*
0.72 0.67 0.86 0.41
1.21* 2.07* 1.77* 2.39*
400–450
Results are shown for the 350–400 and the 400–450 ms window. * pb0.01 (increase in type I error was controlled for every contrast by means of the Hochberg step-up approach).
52
M.I. Nu´n˜ez-Pen˜a, M.L. Honrubia-Serrano / International Journal of Psychophysiology 56 (2005) 45–54
Table 3 Mean (in AV) and standard error (in brackets) obtained for the different ending words in the two tasks 500–600
Typical Atypical Nonmember
600–700
Response task
Reading task
1.51 (0.59) 0.83 (0.74) 3.78 (0.90)
0.42 (0.70) 1.00 (0.48) 0.90 (0.44)
Dif. 1.09 0.17 2.88*
700–800
Response task
Reading task
0.91 (0.44) 1.05 (0.61) 5.63 (0.89)
0.37 (0.76) 1.45 (0.60) 1.66 (0.49)
Dif. 0.54 0.40 3.97*
Response task
Reading task
0.00 (0.47) 0.92 (0.46) 4.9 (1.03)
0.69 (0.75) 1.55 (0.63) 2.38 (0.57)
Dif. 0.69 0.63 2.52*
Results are shown for the 500–600, 600–700 and 700–800 ms windows. Dif.: difference (in AV). * pb0.01 (increase in type I error was controlled for every contrast by means of the Hochberg step-up approach).
between 0.17 and 1.09 AV were reported for the member ending words. Another interaction that reached statistical significance in the 500–600, the 600–700 and the 700–800 ms intervals was the Type of task Electrode interaction: F(6,90)=2.71, p=0.05, e=0.52, g 2=0.15 in the 600–700 ms window, F(6,90)=3.02, p=0.010, e=0.66, g 2 =0.17 in the 700–800 ms window and F(6,90)=3.12, p=0.008, e=0.67, g 2=0.17 in the 700– 800 ms window. Simple effect analyses revealed that differences between tasks were present in centroparietal sites. Statistically significant differences between the tasks were found at C4, P3, P4 and Pz in the 500–600 ms window and at C3, C4, P3, P4 and Pz in the 600–700 ms window. As for the other window, 700–800 ms, differences between tasks were only found in P3.
4. Discussion The primary goal of the present study was to examine whether the N400 amplitude can be considered an index of the degree of category exemplar associative strength. To this end, a semantic category was defined by means of a series of words, and the ERP response was measured after presenting a word that could fit the category to a greater or lesser degree. As far as the relationship between N400 and semantic incongruity is concerned, our results are consistent with those obtained in previous studies (Polich, 1985); however, no difference on the N400 amplitude between atypical members and nonmembers of the category was found. As has already been mentioned, the present experiment provides further evidence of an N400 effect related to clear semantic mismatches between
a word and the previous context. Indeed, a negative component was elicited whenever the word presented did not fit in the previously described semantic category. This component was present in the 350– 450 ms interval and showed a centro-parietal distribution. Therefore, it is concluded that this negative peak is the well-known semantic N400 component. Although a gradation of the N400 amplitude in relation to the degree of category exemplar associative strength has been hypothesized if the semantic features of the category were activated, no evidence of it was found in our study. Instead, a N400 component of similar amplitude was elicited by nonmembers and by atypical members of the category. In other words, the N400 component was elicited whenever the more preferred exemplar for that category was not presented. Our failure to find a category exemplar associative strength effect on the N400 amplitude can be explained in the following way. When a series of words are used as a semantic context, the best exemplar of the category is activated not the semantic features of the category. So, a N400 component of similar amplitude was present when an atypical member or a nonmember of the category was presented. Some studies have shown a semantic priming effect when pairs of words are presented; N400 amplitude to targets is attenuated for semantically related word pairs compared to unrelated word pairs (Bentin et al., 1985; Holcomb and Neville, 1990). In our experiment, the series of words were presented in order from low to high total frequency so the sixth word was always a high frequency exemplar for the category. As the atypical members were low-frequency exemplars, the semantic relationship between the sixth and
M.I. Nu´n˜ez-Pen˜a, M.L. Honrubia-Serrano / International Journal of Psychophysiology 56 (2005) 45–54
seventh words was usually low. This may explain the fact that there were no N400 amplitude differences between nonmembers and atypical members of the category. Our second aim was to control the possible decision effect over the N400 component. To this end, subjects were asked to perform two different tasks: (a) to decide on the congruity or incongruity of the word with regard to a previously described category and (b) to read the words for comprehension. The main conclusion that can be drawn from the results is that the decision task does not affect the N400 component. Significantly, we observed a category-based effect on the N400 although the categorical relationship between words was not relevant to the comprehension task (in other words, additional attention to semantic relationships had no effect on the amplitude of the N400). This result confirms those obtained in other studies where a N400 component is elicited although it was not necessary to pay attention to the semantic congruity of the stimuli (Bentin et al., 1993). The main conclusions of this study can be summarized as follows: first, when a semantic category is established by means of a series of words, the best exemplar of this category is activated; second, the N400 amplitude seems to be sensitive to category membership although no specific attention to the semantic relationship between words was required. Acknowledgments We thank Antonio Solanas, Carles Escera and Nuria Sebastia´n for their helpful comments on an earlier draft of this paper. References Alameda, J.R., Cuetos, F., 1995. Diccionario de Frecuencias de Las Unidades Lingqı´sticas del Castellano. Servicio de Publicaciones de la Universidad de Oviedo, Oviedo. Batting, W.F., Montague, W.E., 1969. Category norms for verbal items in 56 categories: a replication and extension of the Connecticut category norms. J. Exp. Psychol. Monogr. 80 (3), 1 – 46. Bentin, S., McCarthy, G., Wood, C.C., 1985. Event-related potentials associated with semantic priming. Electroencephalogr. Clin. Neurophysiol. 60, 343 – 355.
53
Bentin, S., Kutas, M., Hillyard, S., 1993. Electrophysiological evidence for task effects on semantic priming in auditory word processing. Psychophysiology 30, 161 – 169. Brown, C., Hagoort, P., 2000. On the electrophysiology of language comprehension: implications for the human language system. In: Crocker, M.W., Pickering, M., Clifton, C. (Eds.), Architectures and Mechanisms for Language Processing. Cambridge University Press, Cambridge, pp. 213 – 237. Cohen, J., 1988. Statistical Power Analysis for the Behavioral Sciences. Lawrence Erlbaum Associates, Hillsdale, NJ. Federmeier, K.D., Kutas, M., 1999. A rose by any other name: longterm memory structure and sentence processing. J. Mem. Lang. 41, 469 – 495. Fischler, I., Bloom, P.A., Childers, D.G., Roucos, S.E., Perry, N.W., 1983. Brain potentials related to stages of sentence verification. Psychophysiology 20, 400 – 409. Fujihara, N., Nageishi, Y., Koyama, S., Nakajima, Y., 1998. Electrophysiological evidence for the typicality effect of human cognitive categorization. Int. J. Psychophysiol. 29, 65 – 75. Geisser, S., Grenhouse, S.W., 1958. An extension of Box’s results on the use of the F distribution in multivariate analysis. Ann. Math. Stat. 29, 885 – 891. Hays, W.L., 1994. Statistics. 5th ed. Holt, Rinehart & Winston, New York. Heinze, H.J., Muente, T.F., Kutas, M., 1998. Context effects in a category verification task as assessed by event-related brain potential (ERP) measures. Biol. Psychol. 47, 121 – 135. Hochberg, Y., 1988. A sharper Bonferroni procedure for multiple tests of significance. Biometrika 75 (4), 800 – 802. Holcomb, P.J., Neville, H.J., 1990. Auditory and visual semantic priming in lexical decision: a comparison using event-related brain potentials. Lang. Cogn. Processes 5, 281 – 312. Keselman, H.J., 1998. Testing treatment effects in repeated measures designs: an update for psychophysiological researchers. Psychophysiology 35, 470 – 478. Keselman, H.J., Keselman, J.C., 1988. Comparing repeated measures means in factorial designs. Psychophysiology 25 (5), 612 – 618. Keselman, H.J., Keselman, J.C., 1993. Analysis of repeated measurements. In: Edwards, L.K. (Ed.), Applied Analysis of Variance in Behavioral Science. Marcel Dekker, New York, pp. 105 – 145. Kirk, R.E., 1996. Practical significance: a concept whose time has come. Educ. Psychol. Meas. 56 (5), 746 – 759. Kounios, J., Holcomb, P., 1992. Structure and process in semantic memory: evidence from event-related brain potentials and reaction times. J. Exp. Psychol. Gen. 121, 460 – 480. Kutas, M., Federmeier, K.D., 2000. Electrophysiology reveals semantic memory use in language comprehension. Trends Cogn. Sci. 12 (4), 463 – 470. Kutas, M., Hillyard, S.A., 1980. Reading senseless sentences: brain potentials reflect semantic incongruity. Science 207, 203 – 205. Kutas, M., Hillyard, S.A., 1984. Brain potentials during reading reflect word expectancy and semantic association. Nature 307, 161 – 163.
54
M.I. Nu´n˜ez-Pen˜a, M.L. Honrubia-Serrano / International Journal of Psychophysiology 56 (2005) 45–54
Kutas, M., Van Petten, C., 1994. Psycholinguistics electrified: event-related potential investigations. In: Gernsbacher, M.A. (Ed.), Handbook of Psycholinguistics. Academic Press, San Diego, CA, pp. 83 – 143. Kutas, M., Lindamood, T., Hillyard, S.A., 1984. Word expectancy and event-related brain potentials during sentence processing. In: Kornblum, S., Roquin, J. (Eds.), Preparatory States and Processes. Erlbaum, Hillsdale, NJ, pp. 217 – 238.
Polich, J., 1985. Semantic categorization and event-related potentials. Brain Lang. 26, 304 – 321. Soto, P., Sebastia´ n, M.V., Garcı´a, E., del Amo, T., 1982. Categorizacio´n y datos normativos en espan˜a. Coleccio´n de monografı´as del ICE, Madrid.
N400 and category exemplar associative strengthB M. Isabel Nu´n˜ez-Pen˜a*, M. Luisa Honrubia-Serrano University of Barcelona, Department of Behavioral Sciences Methods, Faculty of Psychology, Passeig Vall d’Hebron, 171, 08035 Barcelona, Spain Received 3 August 2004; received in revised form 14 September 2004; accepted 28 September 2004
Abstract The main purpose of this study was to examine if the N400 amplitude can be considered an index of category exemplar strength of association in a semantic categorization task. Series of six words from a particular semantic category were used as context, and the strength of association between the seventh word and the category was manipulated. Moreover, subjects were asked to perform two different tasks: one consisting of reading the words for comprehension and the other of making a decision regarding the congruence of the ending word. Results showed that the N400 component was elicited by whichever word that was not the best exemplar for the category (atypical members and nonmember of the category). No interaction between type of ending word and type of task was found around 400 ms poststimuli although the P3b component was present for infrequent stimuli in the decision task. It is concluded that the N400 amplitude is sensitive to category membership although no specific attention to the semantic relationship between words was required. D 2004 Elsevier B.V. All rights reserved. Keywords: N400; P3b; Semantic categories; ERPs
1. Introduction The interest in the nature of the processes underlying the N400 component has grown significantly since it was first described by Kutas and Hillyard (1980). These authors contrasted semantically conB Support: This research was supported by grants FI-PG/951.191 from the Generalitat de Catalunya, and BS02003-02440 from the Spanish Ministry of Science and Technology. * Corresponding author. Tel.: +34 93 312 58 53; fax: +34 93 402 13 59. E-mail address: [email protected] (M.I. Nu´n˜ez-Pen˜a).
0167-8760/$ - see front matter D 2004 Elsevier B.V. All rights reserved. doi:10.1016/j.ijpsycho.2004.09.006
gruent ending words with semantically incongruent ending words in sentences and found that incongruent final words elicited a negative wave. This negativity reached its largest amplitude around 400 ms and showed a posterior distribution. This newly discovered component was called N400. Since then, many studies have been conducted to find out more about the nature of the N400 component (for a review, see Brown and Hagoort, 2000; Kutas and Federmeier, 2000; Kutas and Van Petten, 1994). Although N400 was first described as ban electrophysiological sign of the dreprocessingT of semantically anomalous informationQ (Kutas and
46
M.I. Nu´n˜ez-Pen˜a, M.L. Honrubia-Serrano / International Journal of Psychophysiology 56 (2005) 45–54
Hillyard, 1980, p. 203), it was later considered an bindex of the degree of semantic priming or activation that a word receives from the prior contextQ (Kutas et al., 1984, p. 216). In fact, Kutas and Hillyard (1984) found that the N400 amplitude was inversely related to the cloze probability of a word in a sentence (the cloze probability of a word is the frequency in which this word is produced as a completion of a particular sentence). The easier to integrate the word in the sentence, the smaller the N400 amplitude. This result showed that N400 was not merely an index of semantic incongruity because it was also elicited by semantically correct but less preferred endings. The N400 component is also sensitive to category membership. Several studies have shown that it is elicited whenever the presented word does not fit in a previously established semantic category. To date, N400 sensitivity to category membership has been observed both in sentences and in lists or pairs of words. Polich (1985) presented series of words belonging to the same semantic category and reported a negative wave whenever the ending word was not an exemplar of this category—in this case, the semantic category was established by the presentation of a series of words. Category effects on the N400 have also been observed during the performance of sentence verification tasks. In sentences like bA robin is a birdQ (Fischler et al., 1983) or bAll collies are dogs,Q (Kounios and Holcomb, 1992) the N400 component was elicited whenever the exemplar did not fit the category. In spite of the fact that true and false sentences were presented, the N400 component was independent of the truth–value of the sentence being modulated instead by the relatedness of the subject and predicate terms. The N400 component seems to be sensitive not only to category membership but also to the category exemplar associative strength. Heinze et al. (1998) presented subjects with pairs of words, the first of which was the name of a semantic category and the second a typical member, an atypical member or a nonmember of the category. These authors found that the N400 component was elicited for both atypical and nonmember words although the N400 amplitude was larger for the nonmembers. Similar results were obtained by Fujihara et al. (1998). Taken together, all
these results suggest that the semantic typicality of an exemplar modulates the N400 amplitude. Semantic categorization effects on the N400 have also been observed in the context of sentences. Federmeier and Kutas (1999) presented sentences that ended with the expected item, an unexpected item from the same category as the expected exemplar (within-category violations) or an unexpected item from a different category than the expected exemplar (between-category violations). The expected exemplar elicited a late positivity, but both violations elicited a different N400 response: a larger N400 amplitude for between- than for within-category violations (a similar result had been reported by Kutas et al., 1984). Federmeier and Kutas claimed that bsemantic features of the category exemplar (not necessarily a specific lexical item) most likely to complete the target sentence are activated prior to the presentation of the actual sentence-final word. When the prediction is incorrect and the expectancy is not met, the data are characterized by increased N400 activity relative to when the prediction is correctQ (Federmeier and Kutas, 1999, p.487). This explanation also accounts for the results by Heinze et al. (1998). Summarizing, the N400 seems to be a useful component to study semantic categorization effects because its amplitude is modulated by the degree of semantic relatedness between words. The aim of the present experiment was to study if the N400 amplitude is influenced by the degree of category exemplar associative strength. To this end, series of words belonging to the same category were used as context; the ending word could be a typical member, an atypical member or a nonmember of the category previously described. To our knowledge, this is the first study of any kind to examine the degree of category exemplar associative strength by using series of words to establish the semantic category. Cloze probability experiments have shown that the N400 amplitude seems to be an index of degree of semantic integration of a word in the previous context of a sentence. On the other hand, Federmeier and Kutas (1999) claimed that the N400 amplitude could be related to the activation of the semantic features of upcoming words prior to their occurrence. Our goal was to study if, given a series of words that define a semantic category, the amplitude of the N400 would be affected by the
M.I. Nu´n˜ez-Pen˜a, M.L. Honrubia-Serrano / International Journal of Psychophysiology 56 (2005) 45–54
category exemplar associative strength. If semantic features of the category are activated, then a N400 amplitude modulation is expected depending on the strength of association between the word and the category; the more incongruent the ending word presented with regard to the category, the larger the N400 amplitude. If a typical member of the category is activated, then it is expected to find an N400 of similar amplitude for atypical members and nonmembers of the category.
2. Materials and methods 2.1. Subjects Sixteen volunteers (four men; one left handed; mean age=24.8 years, standard error=1.5) participated in the experiment. They were native speakers of Spanish and had normal or corrected to normal vision. All gave informed consent to participate in the experiment. 2.2. Stimuli The stimuli consisted of 168 series of seven words. The first six words belonged to the same semantic category, and the seventh could be a typical member, an atypical member or a nonmember of the category previously established. These words were selected from the material produced by Soto et al. (1982). The procedure that Soto et al. followed for their material was similar to that used by Batting and Montague (1969). It consisted of asking the subjects to write down words belonging to the same semantic category in a notebook. Then total and partial frequencies were recorded for every word. Total frequency was computed by adding the number of times that a word was written in whatever position. Partial frequency, however, was computed by adding the number of times that a word was written in every position—first to seventh positions. The series was constructed as follows. In the first place, we selected words in second to seventh total frequency position for each category (compound words were never selected); these words were presented sequentially from lower to higher frequency and constituted the series. In the second place, the
47
ending word in the series was defined in the following way. The typical member was the word with highest total and partial frequency in a category. Usually, these two criteria matched; nevertheless, when this was not the case, the word with highest partial frequency was selected. To select the atypical member, the following two conditions had to occur: (1) the result of dividing the total frequency of the word selected as atypical member by the total frequency of the word that occupied the seventh position had to be around 0.2—words selected for this condition had a mean of 0.2, with a maximum of 0.3 and a minimum of 0.13; (2) the word had to present its higher partial frequency from third position. As for the nonmembers, these words were selected from the Alameda and Cuetos (1995) Spanish frequency dictionary following these criteria: (1) they were nouns; (2) they had high use frequency (mean=564.3; standard error=27.9); (3) they were nonmembers of the semantic category previously described; (4) they had not been selected as a part or ending of the series in the experiment. The experiment was controlled by the STIM 2.0 program (NeuroScan, Herndon, VA). Words were presented in uppercase and white color letters on a black background. The stimuli subtended a visual angle of 1.238 vertically and a mean of 4.98 horizontally. 2.3. Procedure Participants were tested in two experimental sessions carried out in separate days. The material presented in the two experimental sessions was the same, but participants were given different tasks in each. One task consisted of pressing one button on the mouse if the ending word was congruent with the previous series and the other button if the ending word was incongruent—the assignment of congruent and incongruent response to left or right button was counterbalanced across subjects (henceforth, the response task). In the other task, however, participants were instructed to read the words silently for comprehension because they had to answer a questionnaire about the stimuli at the end of the recording (henceforth, the reading task). The purpose of working with these two sessions was to control for possible decision effects on the ERPs. Each session lasted
48
M.I. Nu´n˜ez-Pen˜a, M.L. Honrubia-Serrano / International Journal of Psychophysiology 56 (2005) 45–54
about 2 h, and the order of sessions was counterbalanced across subjects. A minimum period of 30 days was left between sessions to minimize memory effects. Trials in both sessions consisted of series of seven sequentially presented words. A thousand ms after the ending word, an asterisk appeared on the screen giving the subject the following information: (1) the series had finished, (2) blinking was allowed, and (3) a response had to be emitted. Regarding this last point, motor response was only compulsory in the response task. Fig. 1 shows the outline of a trial. In both sessions, subjects sat in a comfortable armchair approximately 60 cm from a monitor. The session began with a short practice period to acclimatize them to the experimental conditions, consisting of several trials like those presented during the recording period. Participants were told to press one button if the ending word belonged to the previous semantic category and the other button if the ending word did not. They were then given feedback on the correctness of the response. Because our main aim was to ensure that subjects discriminated between experimental conditions, the training period did not finish until one of these learning criteria was met: (1) the participant had correctly answered the first 10 consecutive trials, or (2) the subject correctly answered 90% of trials. There was a 30-s rest after every 20 trials. When the training period was over, the recording period started. Subjects were instructed to relax and to keep their eyes on the screen. They were also told to avoid blinking; if they needed to blink, they should try to wait until an asterisk or a rest message appeared on the screen. Participants were presented with 126 trials in each session—42 of each type of stimulus— which were organized in 14 blocks of nine trials. Three trials of each type of stimulus were randomly presented in each block with the restriction that no more than three trials of the same type could be presented consecutively. Moreover, the same category was never presented twice inside a block. Participants
were given a 30-s rest after each block to allow them a time to blink. Words were presented for 500 ms with an interstimulus interval of 1000 ms. The intertrial period was 2000 ms. At the end of the recording in the reading task, participants were asked to answer both a memory and a recognition test. In the memory test, participants had to write down on a sheet of paper the name of the semantic categories that had been presented during the recording session; they were instructed to write as many categories as they could remember. In the recognition test, a 42-word list was presented, and subjects were requested to specify for each word if it had been previously presented during the recording session—32 of the 42 items in the word list had been previously presented during the recording session. The responses of these two tests allowed us to assess whether a participant had paid sufficient attention to the material. 2.4. EEG recording Recording and analysis of ERPs were performed with the SCAN 3.0 software (NeuroScan, Herndon, VA). EEG was recorded from seven tin electrodes mounted in a commercial electro-cap (Electro-Cap International, Eaton, OH) and positioned according to the 10–20 International System: C3, C4, P3, P4, Cz, Fz and Pz. Connected ear lobes served as reference, and an equidistant point between Fpz and Fz was used as location of the ground electrode. For monitoring eye movement and blinks, four additional electrodes were used: two for the VEOG recording placed above and below the right eye and two for the HEOG recording placed at the two external canthi. EEG and EOG channels were continuously digitized at a rate of 250 Hz by a SynAmpTM amplifier (5083 model, NeuroScan, Herndon, VA). A band pass filter was set from 0.05 to 30 Hz, and a notch filter of 50 Hz was used. Electrode impedance was always kept below 5 kV.
Fig. 1. Scheme of a nonmember ending trial.
M.I. Nu´n˜ez-Pen˜a, M.L. Honrubia-Serrano / International Journal of Psychophysiology 56 (2005) 45–54
2.5. Statistical analysis The statistical analysis was done in three steps: (1) analyses of the memory and the recognition test, (2) analyses of the behavioral data and (3) analyses of the electrophysiological data. Memory and recognition test were analyzed by means of descriptive statistics for the percentage of correctly remembered categories and the percentage of correctly recognized words. Behavioral data analyses could only be performed in the response task. Because participants had to give their motor response after the asterisk and not after the ending word of the series, reaction time was not a useful variable; therefore, number of errors was the only behavioral variable analyzed. Pressing the wrong button was considered an error; however, nonpressing and responses given outside the interval from 100 to 1900 ms were not counted as errors. Nonparametric statistics were used for the number of error analyses: first, the Friedman test was executed for the variable Type of ending word (typical member, atypical member and nonmember); second, the Wilcoxon test was performed for paired contrasts. The analysis of the electrophysiological responses was carried out according to this procedure. First, epochs were averaged for every subject in each experimental condition relative to a prestimulus baseline that comprised the 100 ms of activity preceding the epoch of interest. Trials with artifacts (voltage exceeded F50 Av in any channel) and those with response errors were excluded from the ERP averages. Second, ERP data were analyzed by computing the mean amplitude in 50 ms windows from 250 to 500 ms poststimuli and in 100 ms windows from 500 to 800 ms poststimuli. We analyzed 50 ms windows between 250–500 ms to identify early negativities and 100 ms windows between 500–800 ms to look for late positivities. The purpose of working with all these windows was to extract a detailed analysis of the waves. A 237 repeated measures analysis of variance (ANOVAs) was performed on the amplitude means, taking as independent variables Type of task (response or reading), Type of ending word (typical member, atypical member and nonmember) and Electrode (C3, C4, P3, P4, Fz, Cz and Pz). Repeated measures ANOVAs were performed with the Greenhouse–
49
Geisser correction (Geisser and Greenhouse, 1958) for sphericity departures, which was applied when appropriate; we reported the uncorrected degrees of freedom, the e value, the probability level following correction and the g 2 effect size index (Cohen, 1988; Hays, 1994; Kirk, 1996). Whenever a main effect reached statistical significance—p-value less than or equal to 0.05—, t-tests contrasts were calculated (Keselman and Keselman, 1988; Keselman and Keselman, 1993); to analyze statistically significant interaction, simple effects were calculated. In both cases, the Hochberg step-up approach was used to control for the increase in type I error (Hochberg, 1988; Keselman, 1998).
3. Results 3.1. Memory and recognition test Descriptive analyses of the memory and the recognition tests showed the following results. The mean of the percentage of correctly remembered categories was 62.8 (standard deviation=13.1; range 42.8–80.9). As for the percentage of correctly recognized words, the mean was 71.2 (standard deviation=10.3; range 57.1–88.1). These results allow us to conclude that participants understood the experimenter’s instructions and paid sufficient attention to the material during the reading task. 3.2. Behavioral data An ending word effect for the number of errors was obtained in the reading task (v 2[2, N=16]=10.43, p=0.005). Differences were found between typical and atypical members (Z= 2.99, p=0.003); there were more errors for atypical members. However, there were no statistically significant differences between nonmembers and both types of members. The median number of errors was 0, 4 and 1.5 for typical, atypical and nonmembers, respectively. 3.3. Electrophysiological data Grand average ERPs from all recording sites are shown in Fig. 2. In the response task, a negativity peak about 400 ms was present in atypical member
50
M.I. Nu´n˜ez-Pen˜a, M.L. Honrubia-Serrano / International Journal of Psychophysiology 56 (2005) 45–54
Fig. 2. Top: grand average ERPs for each type of ending word in the response task. Bottom: grand average ERPs for each type of ending word in the reading task.
and nonmember ending words. In the nonmember ending words, this negativity was followed by a positive wave peaking around 600 ms poststimulus and larger in centro-parietal sites. The pattern of voltage in the reading task was very similar to that described for the response task. However, a difference for the nonmember endings was present; in the reading task, the negative peak was not followed by the late positive component. In fact, no difference
seemed to exist in the reading task between the three types of ending word from 500 ms. 3.3.1. 200–500 ms interval A main effect of the Type of ending word was obtained in the 350–400 ms window ( F[2,30]=5.91, p=0.007, e=0.91, g 2=0.28) and the 400–450 ms window ( F[2,30]=8.35, p=0.001, e=0.98, g 2=0.35). Paired contrasts were conducted for both windows,
M.I. Nu´n˜ez-Pen˜a, M.L. Honrubia-Serrano / International Journal of Psychophysiology 56 (2005) 45–54 Table 1 Mean (in AV) and standard error (in brackets) for the different ending words presented 350–400 Typical member Atypical member Nonmember
400–450
1.08 (0.41) 0.13 (0.42) 0.61 (0.57)
0.49 (0.44) 0.84 (0.56) 1.41 (0.59)
Results are shown for the 350–400 and the 400–450 ms windows.
and amplitude differences were found between typical and atypical members and between typical members and nonmembers ( p-valuesb0.01 in each contrast); mean amplitude was more negative for atypical members and nonmembers than for members of the category at these windows. Mean amplitude for the different ending types is shown in Table 1. The interaction Series ending type Electrode reached statistical significance in the 350–400 ( F[12,180]=2.45, p=0.005, e=0.28, g 2=0.14) and the 400–450 ms window ( F[12,180]=2.92, p=0.001, e=0.30, g 2=0.16). Simple effect analyses showed that differences for the type of series ending were statistically significant in C4, Cz, P3, P4 and Pz. Therefore, the effect of the series ending type was only revealed in the centro-parietal sites. Paired contrasts were performed for the series ending type at each location, and the following results were found. In the 400–450 ms window, differences were found between typical and atypical members and between typical members and nonmembers ( p-valuesb0.01 in each contrast) at C4, Cz, P3, P4 and Pz. Again, amplitude was more negative for atypical members and nonmembers than for typical members of the category. In the 350–400 ms window, the same differences between typical members and nonmembers were found at C4, Cz, P3, P4 and Pz ( p-valuesb0.01 in each contrast). However,
51
differences between typical and atypical members were only present at P3 and Pz. Table 2 shows mean differences for the previous contrasts. The variable Type of task showed neither main effect nor interactions in the time windows in the 250– 500 ms interval. Therefore, it is concluded that Type of task had no effect on these time windows. 3.3.2. 500–800 ms interval The main effect of Type of ending word was reported in the 500–600 ms window ( F[2,30]=4.72, p=0.017, e=0.96, g 2=0.23), in the 600–700 ms window ( F[2,30]=11.88, pb0.001, e=0.98, g 2=0.42) and in the 700–800 ms window ( F[2,30]=11.79, pb0.001, e=0.88, g 2=0.44). Paired contrasts revealed that nonmember endings elicited a more positive voltage than the member endings in the interval between 500 and 800 ms poststimulus ( p-valuesb0.01). The Type of ending word effect was modulated by the Type of task—as the analysis of the Type of ending word Type of task interaction shows—in the 500–600 ms window ( F[2,30]=3.57, p=0.04, e=0.81, g 2=0.19), in the 600–700 ms window ( F[2,30]=6.87, p=0.003, e=0.78, g 2=0.31) and in the 700–800 ms window ( F[2,30]=4.27, p=0.023, e=0.91, g 2=0.22). Simple effect analyses showed that differences between reading and response tasks were present in nonmember endings but not in both types of member words. While there was no difference between response and reading task for member words, nonmember endings elicited a more positive voltage in the response task than in the reading task. Mean amplitude for each ending word in the three windows can be seen in Table 3. Differences between tasks of 2.88, 3.97 and 2.52 AV for the 500– 600, the 600–700 and the 700–800 ms windows, respectively, were reported when the ending word was a nonmember, whereas differences between tasks
Table 2 Mean differences (in AV) obtained after subtracting amplitudes for nonmembers and atypical members from amplitudes for typical members at every electrode Time (ms)
Contrast
C3
C4
P3
P4
Cz
Fz
Pz
350–400
Typical–atypical Typical–nonmember Typical–atypical Typical–nonmember
0.14 1.12 0.49 1.36
1.07 1.47* 1.41* 1.83*
1.35* 2.85* 1.91* 3.16*
1.01 1.81* 1.49* 2.26*
1.12 1.86* 1.38* 1.86*
0.72 0.67 0.86 0.41
1.21* 2.07* 1.77* 2.39*
400–450
Results are shown for the 350–400 and the 400–450 ms window. * pb0.01 (increase in type I error was controlled for every contrast by means of the Hochberg step-up approach).
52
M.I. Nu´n˜ez-Pen˜a, M.L. Honrubia-Serrano / International Journal of Psychophysiology 56 (2005) 45–54
Table 3 Mean (in AV) and standard error (in brackets) obtained for the different ending words in the two tasks 500–600
Typical Atypical Nonmember
600–700
Response task
Reading task
1.51 (0.59) 0.83 (0.74) 3.78 (0.90)
0.42 (0.70) 1.00 (0.48) 0.90 (0.44)
Dif. 1.09 0.17 2.88*
700–800
Response task
Reading task
0.91 (0.44) 1.05 (0.61) 5.63 (0.89)
0.37 (0.76) 1.45 (0.60) 1.66 (0.49)
Dif. 0.54 0.40 3.97*
Response task
Reading task
0.00 (0.47) 0.92 (0.46) 4.9 (1.03)
0.69 (0.75) 1.55 (0.63) 2.38 (0.57)
Dif. 0.69 0.63 2.52*
Results are shown for the 500–600, 600–700 and 700–800 ms windows. Dif.: difference (in AV). * pb0.01 (increase in type I error was controlled for every contrast by means of the Hochberg step-up approach).
between 0.17 and 1.09 AV were reported for the member ending words. Another interaction that reached statistical significance in the 500–600, the 600–700 and the 700–800 ms intervals was the Type of task Electrode interaction: F(6,90)=2.71, p=0.05, e=0.52, g 2=0.15 in the 600–700 ms window, F(6,90)=3.02, p=0.010, e=0.66, g 2 =0.17 in the 700–800 ms window and F(6,90)=3.12, p=0.008, e=0.67, g 2=0.17 in the 700– 800 ms window. Simple effect analyses revealed that differences between tasks were present in centroparietal sites. Statistically significant differences between the tasks were found at C4, P3, P4 and Pz in the 500–600 ms window and at C3, C4, P3, P4 and Pz in the 600–700 ms window. As for the other window, 700–800 ms, differences between tasks were only found in P3.
4. Discussion The primary goal of the present study was to examine whether the N400 amplitude can be considered an index of the degree of category exemplar associative strength. To this end, a semantic category was defined by means of a series of words, and the ERP response was measured after presenting a word that could fit the category to a greater or lesser degree. As far as the relationship between N400 and semantic incongruity is concerned, our results are consistent with those obtained in previous studies (Polich, 1985); however, no difference on the N400 amplitude between atypical members and nonmembers of the category was found. As has already been mentioned, the present experiment provides further evidence of an N400 effect related to clear semantic mismatches between
a word and the previous context. Indeed, a negative component was elicited whenever the word presented did not fit in the previously described semantic category. This component was present in the 350– 450 ms interval and showed a centro-parietal distribution. Therefore, it is concluded that this negative peak is the well-known semantic N400 component. Although a gradation of the N400 amplitude in relation to the degree of category exemplar associative strength has been hypothesized if the semantic features of the category were activated, no evidence of it was found in our study. Instead, a N400 component of similar amplitude was elicited by nonmembers and by atypical members of the category. In other words, the N400 component was elicited whenever the more preferred exemplar for that category was not presented. Our failure to find a category exemplar associative strength effect on the N400 amplitude can be explained in the following way. When a series of words are used as a semantic context, the best exemplar of the category is activated not the semantic features of the category. So, a N400 component of similar amplitude was present when an atypical member or a nonmember of the category was presented. Some studies have shown a semantic priming effect when pairs of words are presented; N400 amplitude to targets is attenuated for semantically related word pairs compared to unrelated word pairs (Bentin et al., 1985; Holcomb and Neville, 1990). In our experiment, the series of words were presented in order from low to high total frequency so the sixth word was always a high frequency exemplar for the category. As the atypical members were low-frequency exemplars, the semantic relationship between the sixth and
M.I. Nu´n˜ez-Pen˜a, M.L. Honrubia-Serrano / International Journal of Psychophysiology 56 (2005) 45–54
seventh words was usually low. This may explain the fact that there were no N400 amplitude differences between nonmembers and atypical members of the category. Our second aim was to control the possible decision effect over the N400 component. To this end, subjects were asked to perform two different tasks: (a) to decide on the congruity or incongruity of the word with regard to a previously described category and (b) to read the words for comprehension. The main conclusion that can be drawn from the results is that the decision task does not affect the N400 component. Significantly, we observed a category-based effect on the N400 although the categorical relationship between words was not relevant to the comprehension task (in other words, additional attention to semantic relationships had no effect on the amplitude of the N400). This result confirms those obtained in other studies where a N400 component is elicited although it was not necessary to pay attention to the semantic congruity of the stimuli (Bentin et al., 1993). The main conclusions of this study can be summarized as follows: first, when a semantic category is established by means of a series of words, the best exemplar of this category is activated; second, the N400 amplitude seems to be sensitive to category membership although no specific attention to the semantic relationship between words was required. Acknowledgments We thank Antonio Solanas, Carles Escera and Nuria Sebastia´n for their helpful comments on an earlier draft of this paper. References Alameda, J.R., Cuetos, F., 1995. Diccionario de Frecuencias de Las Unidades Lingqı´sticas del Castellano. Servicio de Publicaciones de la Universidad de Oviedo, Oviedo. Batting, W.F., Montague, W.E., 1969. Category norms for verbal items in 56 categories: a replication and extension of the Connecticut category norms. J. Exp. Psychol. Monogr. 80 (3), 1 – 46. Bentin, S., McCarthy, G., Wood, C.C., 1985. Event-related potentials associated with semantic priming. Electroencephalogr. Clin. Neurophysiol. 60, 343 – 355.
53
Bentin, S., Kutas, M., Hillyard, S., 1993. Electrophysiological evidence for task effects on semantic priming in auditory word processing. Psychophysiology 30, 161 – 169. Brown, C., Hagoort, P., 2000. On the electrophysiology of language comprehension: implications for the human language system. In: Crocker, M.W., Pickering, M., Clifton, C. (Eds.), Architectures and Mechanisms for Language Processing. Cambridge University Press, Cambridge, pp. 213 – 237. Cohen, J., 1988. Statistical Power Analysis for the Behavioral Sciences. Lawrence Erlbaum Associates, Hillsdale, NJ. Federmeier, K.D., Kutas, M., 1999. A rose by any other name: longterm memory structure and sentence processing. J. Mem. Lang. 41, 469 – 495. Fischler, I., Bloom, P.A., Childers, D.G., Roucos, S.E., Perry, N.W., 1983. Brain potentials related to stages of sentence verification. Psychophysiology 20, 400 – 409. Fujihara, N., Nageishi, Y., Koyama, S., Nakajima, Y., 1998. Electrophysiological evidence for the typicality effect of human cognitive categorization. Int. J. Psychophysiol. 29, 65 – 75. Geisser, S., Grenhouse, S.W., 1958. An extension of Box’s results on the use of the F distribution in multivariate analysis. Ann. Math. Stat. 29, 885 – 891. Hays, W.L., 1994. Statistics. 5th ed. Holt, Rinehart & Winston, New York. Heinze, H.J., Muente, T.F., Kutas, M., 1998. Context effects in a category verification task as assessed by event-related brain potential (ERP) measures. Biol. Psychol. 47, 121 – 135. Hochberg, Y., 1988. A sharper Bonferroni procedure for multiple tests of significance. Biometrika 75 (4), 800 – 802. Holcomb, P.J., Neville, H.J., 1990. Auditory and visual semantic priming in lexical decision: a comparison using event-related brain potentials. Lang. Cogn. Processes 5, 281 – 312. Keselman, H.J., 1998. Testing treatment effects in repeated measures designs: an update for psychophysiological researchers. Psychophysiology 35, 470 – 478. Keselman, H.J., Keselman, J.C., 1988. Comparing repeated measures means in factorial designs. Psychophysiology 25 (5), 612 – 618. Keselman, H.J., Keselman, J.C., 1993. Analysis of repeated measurements. In: Edwards, L.K. (Ed.), Applied Analysis of Variance in Behavioral Science. Marcel Dekker, New York, pp. 105 – 145. Kirk, R.E., 1996. Practical significance: a concept whose time has come. Educ. Psychol. Meas. 56 (5), 746 – 759. Kounios, J., Holcomb, P., 1992. Structure and process in semantic memory: evidence from event-related brain potentials and reaction times. J. Exp. Psychol. Gen. 121, 460 – 480. Kutas, M., Federmeier, K.D., 2000. Electrophysiology reveals semantic memory use in language comprehension. Trends Cogn. Sci. 12 (4), 463 – 470. Kutas, M., Hillyard, S.A., 1980. Reading senseless sentences: brain potentials reflect semantic incongruity. Science 207, 203 – 205. Kutas, M., Hillyard, S.A., 1984. Brain potentials during reading reflect word expectancy and semantic association. Nature 307, 161 – 163.
54
M.I. Nu´n˜ez-Pen˜a, M.L. Honrubia-Serrano / International Journal of Psychophysiology 56 (2005) 45–54
Kutas, M., Van Petten, C., 1994. Psycholinguistics electrified: event-related potential investigations. In: Gernsbacher, M.A. (Ed.), Handbook of Psycholinguistics. Academic Press, San Diego, CA, pp. 83 – 143. Kutas, M., Lindamood, T., Hillyard, S.A., 1984. Word expectancy and event-related brain potentials during sentence processing. In: Kornblum, S., Roquin, J. (Eds.), Preparatory States and Processes. Erlbaum, Hillsdale, NJ, pp. 217 – 238.
Polich, J., 1985. Semantic categorization and event-related potentials. Brain Lang. 26, 304 – 321. Soto, P., Sebastia´ n, M.V., Garcı´a, E., del Amo, T., 1982. Categorizacio´n y datos normativos en espan˜a. Coleccio´n de monografı´as del ICE, Madrid.