The Effect of Stimulus Repetition on Cortical Magnetic Responses Evoked by Words and Nonwords

The Effect of Stimulus Repetition on Cortical Magnetic Responses Evoked by Words and Nonwords

NeuroImage 14, 118 –128 (2001) doi:10.1006/nimg.2001.0774, available online at http://www.idealibrary.com on The Effect of Stimulus Repetition on Cor...

359KB Sizes 9 Downloads 68 Views

NeuroImage 14, 118 –128 (2001) doi:10.1006/nimg.2001.0774, available online at http://www.idealibrary.com on

The Effect of Stimulus Repetition on Cortical Magnetic Responses Evoked by Words and Nonwords Takahiro Sekiguchi,* ,† Sachiko Koyama,* and Ryusuke Kakigi* *Department of Integrative Physiology, National Institute for Physiological Sciences, Okazaki, Japan; and †Department of General Psychology, Faculty of Human Sciences, Osaka University, Suita, Japan Received September 15, 2000; published online May 11, 2001

Stimulus repetition improves performance and modulates event-related brain potentials in word recognition tasks. We recorded evoked magnetic responses from bilateral temporal sites of the brain to determine the cortical area related to the word repetition effect. Fourteen Japanese volunteers read words or pronounceable nonwords, some of which occurred twice with a lag of eight items. Clear magnetic responses were observed bilaterally. In the left hemisphere, a reduction of the magnetic responses by repetition was observed for words but not for nonwords in the latency range of 300 –500 ms poststimulus. The sources of the responses were estimated to be in the left perisylvian area adjacent to the auditory cortex and the left parietal area. Only the perisylvian source activity showed the reduction by the word repetition. The left perisylvian area was thus suggested to be related to the word repetition effect. The activity in this area might be associated with the lexical memory process. ©

2001 Academic Press

Key Words: magnetoencephalography; word repetition effect; reading; lexical access; N400.

INTRODUCTION Performance in visual word recognition tasks is facilitated by repeated presentation of the same word. For example, the reaction time to a repeated word is shorter than that to the first presentation of the word in both lexical decision (word–nonword decision) (e.g., Scarborough et al., 1977) and naming tasks (e.g., Durso and Johnson, 1979; Feustel et al., 1983). Repetition also makes the response more accurate in tachistoscopic identification tasks (e.g., Feustel et al., 1983; Jacoby, 1983). This phenomenon, referred to as the word repetition effect or repetition priming, has been studied vigorously for over a decade and promises to lead to an understanding of the organization of memory processes (see Tenpenny, 1995, for a review). The word repetition effect has been regarded as a composite effect resulting from at least two factors 1053-8119/01 $35.00 Copyright © 2001 by Academic Press All rights of reproduction in any form reserved.

(Tenpenny, 1995). One is a modification of word representation stored in lexical memory (mental lexicon) (Monsell, 1985; Morton, 1979). In word recognition, sensory input from a word is assumed to activate the mental representation corresponding to it. Since a certain amount of the activation remains in the representation for some period, lexical access for the same word is facilitated on second presentation. The other factor is the retrieval of episodic memory (memory for a specific event) (Feustel et al., 1983; Jacoby, 1983; Salasoo et al., 1985). That is, the recovery of an episodic trace for the prior occurrence of a word helps subjects remember the stimulus identity and the response associated with it. The finding that the repetition effect occurs also for nonwords having no corresponding representations in lexical memory (Feustel et al., 1983; Salasoo et al., 1985) suggests contribution of the episodic factor to the effect. The word repetition effect has been also investigated using scalp recorded event-related brain potentials (ERPs) (see Rugg and Doyle, 1994, for a review). A number of studies have shown that the ERPs elicited by repeated words are more positive-going than those to novel words (e.g., Bentin and Peled, 1990; Rugg, 1987, 1990; Van Petten et al., 1991). This ERP repetition effect starts around 250 ms poststimulus, and persists for a further 300 – 400 ms. It is undiminished as the lag between first and second presentations increases from zero to as many as 19 intervening words (Bentin and Peled, 1990; Nagy and Rugg, 1989). The ERP repetition effect has been surmised to be a result of the modulation of two temporary and spatially overlapping components of ERP. One is the N400 and the other is a late positive component (Rugg, 1990; Van Petten et al., 1991). Although the functional significance of these two components is still uncertain, the N400 may reflect the processes involved in the recognition of words (Van Petten et al., 1991), and the late positive component likely reflects processes involved in the recollection of prior episodes (Rugg et al., 1996; Van Petten et al., 1991). These ERP studies thus also sug-

118

119

WORD REPETITION EFFECT ON MEG

gest that the word repetition effect is related to multiple cognitive functions. In the present study, we recorded magnetoencephalograms (MEGs) to investigate the locus and timing of the word repetition effect in the human brain. MEG is a noninvasive technique for investigating neural activities in the brain. MEG has better spatial resolution than electroencephalogram (EEG) because EEGs are distorted by layers of different impedance, particularly the high impedance of the skull and low impedance of the cerebrospinal fluid. MEGs are less affected by this inhomogeneity. In MEG studies, the weak magnetic fields produced by electric currents flowing in neurons are measured with multi-channel SQUID (Superconducting QUantum Interference Device) magnetometers. The sites in the cerebral cortex activated by stimuli can be estimated from the detected magneticfield distribution. Under favorable conditions, the site can be located with millimeter-range precision (Ha¨ma¨la¨inen et al., 1993). In our previous study, the magnetic responses elicited by repeated words were compared with those to novel words (Sekiguchi et al., 2000). In that study, words were presented visually repeating with a lag of eight items. We found that the magnetic responses evoked by repeated words were smaller than those by first words. This effect was observed bilaterally in the latency range of 250 – 600 ms poststimulus. Source analysis showed that the bilateral perisylvian areas adjacent to the auditory cortex were activated in this latency range, and the activity in these areas was reduced by repetition. This result indicates that the bilateral perisylvian areas were related to the word repetition effect. The aim of the present study was to build on the finding obtained in our previous study, and to elucidate whether the repetition effect found in the perisylvian area is involved in the lexical memory process or in the episodic memory process. For this purpose, we introduced nonwords as stimuli and compared the effect of repetition between the magnetic responses to words and those to nonwords. Nonwords have no mental representations in definition. The presentation of nonword is hence expected to make no change in lexical memory. Therefore, if the magnetic responses are modulated by word repetition but not by nonword repetition, the source area can be considered to be related to the lexical memory process. On the other hand, if the responses are modulated by both the word and nonword repetition, the source area is likely to be related to the retrieval of episodic memory. We recorded the magnetic responses from bilateral temporal sites of the brain. The words and nonwords were presented visually, and some of them appeared twice. The lag of repetition was eight items in order to eliminate the matching of the first and second items in working memory. To avoid recording the brain activity

related to response selection or response output, the subjects were not required to make any overt responses. Attention to the stimuli was ensured by a recognition memory test that was conducted every 31 trials. METHODS Subjects Fourteen healthy Japanese (five females and nine males; age 22 to 40 years) who were staff members or graduate students of the National Institute for Physiological Sciences participated in the experiment. All subjects were right-handed (self-report) and had normal or corrected-to-normal vision. The subjects gave their informed consent, and the study was approved by the Ethical Committee of the National Institute for Physiological Sciences. Stimuli The stimuli were 228 Japanese words and 228 pronounceable nonwords. They were all four letters long and were presented visually in Katakana script (a Japanese syllabic script). The words were nouns usually written in Katakana (e.g., foreign loanwords). Nonwords were made by combining a pair of two-letter strings that had low association value based on a Japanese syllable association norm (Hayashi, 1976), and thereby were orthographically dissimilar to any words. One hundred twenty words and 120 nonwords were presented twice, separated by eight intervening words/ nonwords (i.e., lag 8). The magnetic responses to these items were selectively averaged according to stimulus type (word/nonword) and presentation type (first item/ second item). The remaining 108 words and 108 nonwords were used as filler items and were presented just once to minimize the predictability of where in the presentation sequence the repetition would occur. The magnetic responses to the filler items were not analyzed. An example of the stimulus presentation sequence is shown in Fig. 1. The stimuli were presented in 24 blocks of 31 trials each (744 trials in total). In each block, 5 words, 5 nonwords, 5 repeated words, 5 repeated nonwords, and 11 filler items (including two repetitions, as mentioned below) were presented one by one. They were ordered pseudorandomly across the subjects, keeping the second presentation of the items at the lag eight from their first presentation. No items were repeated across the blocks. In each block, the first two items were always the fillers. Unlike the other filler items, these were presented twice with a lag of two to seven items to prevent the first nine items from having no repetitions. A recognition memory test followed each block. One word or nonword was presented in the test. In 15

120

SEKIGUCHI, KOYAMA, AND KAKIGI

FIG. 1. Example of a stimulus presentation sequence. Words and nonwords were presented in Japanese Katakana script. All stimuli, except for filler items, appeared twice with a lag of eight items (1st and 2nd presentations). The filler items were presented only once.

blocks, an item that appeared in that block was presented (old items). It was one of the repeated items to make the test easy to perform. In nine blocks, a new word or new nonword was presented.

ory test). The experimenter told the subjects whether their answer was correct or not before starting the next block. If necessary, the subjects were allowed to take a rest every four blocks. They were given one block of practice before the recording.

Procedure The subject lay on a bed with his or her right side down in a magnetically shielded room. The shielded room was darkened during the recording. The stimulus presentation was controlled by a computer (PC9801BS2, NEC, Japan), and the stimuli were projected to a screen by a video projector (BARCO3100, BARCODATA, Belgium) set outside the shielded room. The luminance of the stimuli and its background were 230 and 2 cd/m2, respectively. The stimuli were presented for 800 ms with a visual angle of 2.9 ⫻ 0.65°. The intertrial interval varied between 1.0 and 1.4 s. A fixation point (a red dot) was projected at the location corresponding to the center of the stimuli throughout the recording session. To reduce onset responses from the primary visual cortex, a sequence of random-dot patterns (2.9 ⫻ 0.65°), refreshed every 16.7 ms, was presented over the fixation point during the intertrial intervals. The subjects were asked to read the presented words and nonwords silently. They were informed in advance that some items would be repeated at several intervals. At the end of each block, a word or a nonword was presented on the screen. The subjects were required to judge whether this item appeared in the block or not and to report their answer verbally (recognition mem-

MEG Recording and Analysis Magnetic responses were measured with dual 37channel gradiometers (Magnes, Biomagnetic Technologies Inc., U.S.A.). The sensing coils consisted of firstorder axial gradiometers, 20 mm in diameter, with a baseline of 50 mm. The distance between the centers of adjacent coils was 22 mm. The coils were arranged uniformly in concentric circles over a spherical surface. They covered a circle of 14.4 cm in diameter on a spherical surface with a 12-cm radius. The intrinsic noise level of each channel was less than 10fT/公 Hz. The gradiometers were placed over the left and right temporal sites. The magnetic responses were recorded with a 0.1–50 Hz bandpass filter and were recorded digitally from 200 ms before to 800 ms after the stimulus onset at a sampling rate of 520.8 Hz. The magnetic responses were selectively averaged according to stimulus type and presentation type. The mean value of the signals during the 200 ms prior to the stimulus onset was used as the baseline. Epochs with signal variations of larger than 3.0 pT were excluded from the averaging. Vertical and horizontal electro-oculograms (EOGs) were recorded simultaneously to monitor eye movements (bandpass, 0.1–30 Hz; sampling rate, 1024

121

WORD REPETITION EFFECT ON MEG

Hz). Epochs with signal variations of larger than 80 ␮V in the EOG were also excluded from the averaging. The averaged responses included at least 78% of the 120 trials. No digital filter was applied to the averaged responses. If responses in a certain channel were noisy after the averaging, that channel was excluded from analysis. For all subjects, responses from at least 36 channels were actually used for analysis. The strength of the magnetic responses was measured with the standard deviation of the recorded responses across 36 –37 channels (Koyama et al., 2000). A dipolar source below the sensor array produces magnetic fields with high spatial derivatives (the flux is directed outward on one side and inward on the other). The variation between recorded responses would be higher in proportion to the strength of source activity. The standard deviation across channels thus represents the strength of magnetic responses. Compared with root mean square value, which is the index frequently used in MEG studies (e.g., Kaneoke et al., 1997), the standard deviation is resistant to a general drift of the responses produced by a distant source, which often includes sources other than the cerebral cortex. This is because that the distant sources produce relatively uniform fields, and the variation among recorded channels thus is low. The standard deviation is therefore a suitable index of magnetic strength, especially when late responses (300 ms-) are analyzed.

nasion. The positive y-axis was oriented to the left preauricular point, and the positive z-axis was extended toward the vertex. The estimated source locations were overlaid on the subject’s MRIs to identify their anatomical locations. T1-weighted MRIs were obtained using a GE 1.5 T system (slice thickness, 1.0 mm) and a Shimadzu Medical 1.5 T System (slice thickness, 1.5 mm). The same anatomical landmarks used to create the head-based coordinate system were visualized in the MR images by affixing to these points high-contrast liver oil capsules. The common MEG and MRI anatomical landmarks allowed easy transformation of the head-based coordinate system to the MRI. RESULTS Performance in Recognition Memory Test In the recognition memory test, subjects correctly recognized on average 85.7% (SD ⫽ 10.4) of the old words and 80.6% (17.6) of the old nonwords. They made false alarms to 1.8% (6.4) of the new words and 10.0% (12.5) of the new nonwords. Performance was quantified according to the following formula, values derived as correct-recognition rate minus false-alarm rate (Snodgrass and Corwin, 1988). The arcsine-transformed values were subjected to one-way ANOVA. The analysis showed no significant effect between the word and nonword data (F(1, 13) ⫽ 2.26).

Source Analysis During visual word recognition, multiple parts of the brain are expected to be active simultaneously. We thus conducted a multidipole source analysis using a brain electromagnetic source analysis (BESA, Version 2.1m) (Scherg, 1992). BESA has a spatiotemporal modeling approach. It decomposes the recorded responses into a number of discrete neuronal source activities overlapping in time and estimates the strength and timing of each source current. BESA calculates the locations and orientations of dipoles in a spherical head conductor model by an iterative least-squares fit. The goodness-of-fit of the source model was expressed as a percentage of the residual variance (%RV) between the observed fields and the theoretical fields generated by the model. We accepted only the source model in which the %RV was below 10%. Since the BESA has an interactive interface and does not find a source automatically, a single equivalent current dipole modeling (Sarvas, 1987) was fitted for each time point, and the estimated sources were used as hypothesized sources in the BESA. The location of the sources was defined in a headbased coordinate system, set by the bilateral preauricular points and the nasion. The origin of this system was set at the midpoint between preauricular points. The positive x-axis was extended from the origin to the

Magnetic Responses Figure 2 shows representative waveforms of the magnetic responses recorded over the left hemisphere from one subject (S2). Both the words and nonwords elicited clear magnetic responses. Of the responses recorded over the left hemisphere, two components were observed in all the subjects, the 1 M component persisting from approximately 150 to 250 –300 ms poststimulus and the 2 M component ranging from 300 to 500 – 600 ms. Among the responses recorded over the right hemisphere, only the 1 M component was consistently observed across subjects. The effect of repetition was observed in the word data, but not in the nonword data. Figure 3 shows the strength of magnetic responses averaged across 14 subjects. In the word data, the responses to the second words were smaller than those to the first words. This reduction was observed bilaterally in the latency range of 200 –300 ms (1 M) and was observed only in the left hemisphere in the range of 300 –500 ms (2 M). In contrast to the word data, the magnetic responses to the nonwords did not differ between the first and second presentation in either hemisphere or in either component. For statistical analyses, the strength of the magnetic responses measured with the standard deviation was averaged in three latency ranges: 200 –300, 300 –500,

122

SEKIGUCHI, KOYAMA, AND KAKIGI

FIG. 2. Example of magnetic responses evoked by words and nonwords recorded over the left hemisphere in one subject (S2). The waveforms of 37 channels are superimposed. Two sustained components (1 M and 2 M) were identified in the magnetic responses. The repetition reduced the magnetic responses to words but did not affect those to nonwords. The inset figure shows the approximate location of the 37 sensors illustrated by circles.

9.94, P ⬍ 0.01), indicating that the magnetic responses to the second words were smaller than those to the first words in both hemispheres, whereas the ANOVA for the nonword data showed no effect of repetition (F(1, 13) ⬍ 1). In the range of 300 –500 ms, the analysis showed a significant main effect of hemisphere (F(1, 13) ⫽ 11.65, P ⬍ 0.01), indicating that the magnetic responses were larger in the left hemisphere. More importantly, a three-way interaction between stimulus type, repetition, and hemisphere was significant (F(1, 13) ⫽ 6.48, P ⬍ 0.05). The separate two-way ANOVA for the lefthemisphere data showed a significant interaction between stimulus type and repetition (F(1, 13) ⫽ 12.48, P ⬍ 0.01), whereas the ANOVA for the right-hemisphere data showed no effect or interaction. The interaction in the left-hemisphere data indicated that the repetition significantly reduced the magnetic responses to the words (F(1, 13) ⫽ 9.32, P ⬍ 0.01), whereas it did not affect the responses to the nonwords (F(1, 13) ⫽ 2.36). In the range of 500 – 600 ms, while the analysis showed a significant main effect of hemisphere again (F(1, 13) ⫽ 7.77, P ⬍ 0.05), neither the main effect of repetition nor its interaction was significant. In Fig. 3, the 2 M component appears to be sustained for a longer duration for nonwords compared to that for words. This difference was confirmed statistically. In this latency range, the responses to the nonwords were significantly larger than those to the words (F(1, 13) ⫽ 8.12, P ⬍ 0.05), whereas no statistical difference was observed between the word and nonword data in the range of 300 –500 ms.

and 500 – 600 ms poststimulus. 1 The averaged values were analyzed by three-way analyses of variance (ANOVAs), using stimulus type (word/nonword), repetition (first/second), and hemisphere (left/right) as repeated-measure factors. In the range of 200 –300 ms, a main effect of repetition (F(1, 13) ⫽ 4.84, P ⬍ 0.05) and its interaction with stimulus type (F(1, 13) ⫽ 6.64, P ⬍ 0.05) were significant. Separate one-way ANOVA for the word data showed a significant effect of repetition (F(1, 13) ⫽ 1

We compared the results of analyses on standard deviation (SD) values with those of analyses on root mean square (RMS) values and confirmed that very similar results were obtained between these two types of analyses. The reason for reporting the results of the SD analyses is that the SD is more resistant to artifact than the RMS. In fact, ANOVAs showed less clear results in the RMS analysis. Specifically, in the RMS analysis on 300- to 500-ms responses, a threeway interaction did not reach significance (F(1, 13) ⫽ 2.64), although the repetition effect, specific for words, was observed in the left hemisphere but not in the right hemisphere.

FIG. 3. The strength of magnetic responses averaged across 14 subjects. The standard deviations of recorded responses across 36 –37 channels were used as an index of the response strength. In word data, the repetition effect was observed bilaterally in the latency range of 200 –300 ms (1 M) and only in the left hemisphere in the latency range of 300 –500 ms (2 M). In nonword data, the repetition effect was not observed in either hemisphere.

WORD REPETITION EFFECT ON MEG

123

FIG. 4. Estimated source locations for the 2 M component and the time course of their current strength in subjects S1 and S2. The source locations estimated for the first word data were superimposed on the MR image surface renditions. In both subjects, two sources were obtained. One source was located in the left perisylvian area adjacent to the auditory cortex (x, y, and z values of source location were 7, 54, 61 mm in S1 and 16, 46, 61 mm in S2, respectively) and the other source was located in the left parietal area (5, 46, 78 mm in S1 and 4, 48, 84 mm in S2). The vertical lines drawn over the waveforms of the source strength represent the analysis period (300 –500 ms). In this period, the current strength showed a reduction by repetition only in the perisylvian sources for the word data.

Source Analysis The source analysis was conducted for the 1 and 2 M components from the left hemisphere separately. We report here only the results of analysis for the 2 M component because we were not able to obtain consistent results among subjects for the 1 M component. 2 The analysis for the 2 M component was performed for the data in each of the four conditions; first word, second word, first nonword, and second nonword. The analysis period was 300 –500 ms poststimulus, in which the repetition effect was remarkable. The data from one subject (S9) did not show acceptable results in any condition probably due to a low signal-to-noise ratio. We thus describe the results in the remaining 13 subjects below. Figure 4 shows an example of estimated source locations in two subjects (S1 and S2). In all 13 subjects, one source was consistently estimated to be in the left perisylvian area mostly adjacent to the auditory cortex 2 In most subjects, the waveforms of the 1 M component did not show phase reversal across the sensors, suggesting that the source of the magnetic responses was not located under the sensor arrays. Thus, the failure in the source analysis for the 1 M component was probably due to the sensors being placed far from the source area.

(Fig. 4, left). In all subjects except for subject S6, one more source was also found. This source was estimated to be in the inferior part of the left parietal lobe (Fig. 4, right). The obtained source models were similar among the four conditions in all the subjects except for S5. In these 12 subjects, the difference in the source location across the four conditions was less than 18 mm in the perisylvian sources and less than 26 mm in the parietal sources. The x, y, and z values of source locations did not show a significant difference across the conditions. In subject S5, although the nonwords were found to activate different areas compared to words, these source locations were reproduced between first and second presentations and thus deemed reliable. All of the source models showed %RV below 10%, except for first word data in subjects S11 (10.4%) and S13 (14.5%), second word data in S4 (24.1%), S7 (12.2%), and S8 (17.2%), first nonword data in S3 (11.8%), and second nonword data in S6 (10.3%) and S8 (10.4%). These high %RVs were probably due to insufficient magnetic response strength. Figure 4 shows also the estimated current-strength of the obtained sources in subjects S1 and S2. To compare the source strength between the first and second word data, we applied the source model obtained for

124

SEKIGUCHI, KOYAMA, AND KAKIGI

repetition (F(1, 11) ⫽ 16.19, P ⬍ 0.005) and an interaction with stimulus type (F(1, 11) ⫽ 20.04, P ⬍ 0.001). This interaction indicates that the perisylvian activity was significantly reduced (20% on average) by word repetition (F(1, 11) ⫽ 33.79, P ⬍ 0.001), whereas it was not affected by nonword repetition (F(1, 11) ⬍ 1). On the other hand, in the analysis for the strength of the parietal sources, neither a main effect of repetition (F(1, 10) ⫽ 2.27) nor interaction (F(1, 10) ⬍ 1) was significant. These results suggest that the reduction in the left perisylvian activity was responsible for the word repetition effect observed in the 2 M component. FIG. 5. Sources for the magnetic responses to first words and first nonwords collected over 13 subjects. The sources located in the left perisylvian area and those in the parietal area are shown in circles and triangles, respectively (although all of the triangles were not located in the parietal area, we refer to them as the parietal sources for convenience). The gray circles and triangles represent the sources in which the mean strength in the analysis period showed a more than 10% reduction by repetition. These sources were plotted on the schematic brain with the help of individual MRIs. They are projected on the surface of the brain for easy visualization. The perisylvian and parietal sources were located away from the midsagittal plane on average 53 mm (range 45– 66 mm) and 47 mm (range 34 –59 mm), respectively.

the first word data to the second word data. 3 The source strengths for the nonwords were also compared using the models for the first nonword data. In these two subjects, the current strength of the perisylvian sources was reduced by word repetition in the analysis period (300 –500 ms), whereas it was not reduced by nonword repetition. On the other hand, the current strength of the parietal sources did not show a marked reduction either by word or nonword repetition. The same trend was observed in the other subjects. Figure 5 summarizes the results of the source analyses for the first word and first nonword data. The sources estimated to be in the left perisylvian area and the parietal area are plotted as circles and triangles, respectively. Gray circles and gray triangles represent the sources in which the mean strength in the analysis period showed a more than 10% reduction by repetition. It is obvious that the source activity sensitive to the word repetition clustered together in the perisylvian area, and these activities were not sensitive to the nonword repetition. The mean source strength in the analysis period, except for the data from subject S5, underwent twoway ANOVAs using stimulus type (word/nonword) and repetition (first/second) as repeated-measure factors. As a result, the analysis for the strength of the perisylvian sources showed a significant main effect of 3 For word data in subjects S11 and S13, and for nonword data in subject S3, the comparison was made using the model for the second word/nonword data because the model for the first word/nonword data did not show %RV below 10%.

DISCUSSION We recorded evoked magnetic responses from the bilateral temporal sites of the brain to determine the cortical area related to the word repetition effect. Subjects read Japanese words or pronounceable nonwords, some of which occurred twice with a lag of eight items. The left perisylvian area adjacent to the auditory cortex was found to be activated by both the words and nonwords in the latency range of 300 –500 ms poststimulus. This activity was reduced by word repetition, whereas it was not modulated by nonword repetition. This result suggests that the left perisylvian activity is associated with the word repetition effect. The present result is in agreement with our previous finding that the activity in the perisylvian area was reduced by word repetition (Sekiguchi et al., 2000). However, it differs in one important aspect. In the present study, the repetition effect was observed only in the left hemisphere, whereas in our previous study the repetition effect was observed bilaterally. This discrepancy might be due to differences in experimental procedure. While the present study used words and nonwords as stimuli, our previous study used only words. Furthermore, these two studies differ in the manner of stimulus repetition. That is, in the present study the stimulus repetition did not always occur regularly due to the use of unrepeated filler stimuli, whereas in the previous study the word repetition always occurred periodically without the filler stimuli. The word repetition effect in the right hemisphere thus might be influenced by the manner of repetition. Function Associated with the Perisylvian Areas In the present study, the nonword repetition did not modulate the left perisylvian activity. As mentioned in the Introduction, the literature suggests that the word repetition effect is caused by two factors, the modification of word representation and the retrieval of episodic memory. The absence of a nonword repetition effect is well explained by the modification of word representation. In this view, the word repetition effect results from a temporary rise in the activation level in

WORD REPETITION EFFECT ON MEG

a word representation produced by the first presentation of that word. Since nonwords have no corresponding representations in lexical memory, the state of word representations might not altered by the first nonword presentation. The second nonword thus would be processed in the same way as the first nonword, leading to the same cortical activation. On the other hand, episodic factor does not explain the absence of a nonword repetition effect. If the MEG repetition effect observed here was caused by the recollection of the prior occurrence of a first word, the repetition of the nonwords should have also modulated the activity in the left perisylvian area, because the first nonwords might be encoded as episodic memory and be recollected when the second nonwords were presented. The present result is therefore in favor of the view that the MEG repetition effect observed here was induced by the lexical factor and that the activity in the left perisylvian area is associated with the access to word representations. In our previous MEG study, we have found that the activity in the left perisylvian area was reduced by semantic priming for visual words (Koyama et al., 1999). In this study, the former part of a Japanese idiom (prime) was followed by the latter part (target). The magnetic responses to the target preceded by the adequate prime were remarkably smaller than those to the target preceded by the inadequate prime in the latency range of 350 –500 ms poststimulus. In this latency range, the source of the responses was estimated to be in the left perisylvian area, suggesting that this area is related to the semantic priming effect. It is thus suggested that the repetition and semantic priming effects share the same neural substrates in the left perisylvian area. The semantic priming effect is supposed to result from the pre-activation of the target’s mental representation by the spreading activation from mental representations of semantically related words (see Neely, 1991, for a review). Thus, the finding obtained in Koyama et al. (1999) also supports the view that the left perisylvian area is the locus of lexical access. In addition, both word repetition (Rugg, 1990; Van Petten et al., 1991) and semantic priming (Bentin et al., 1985; Koyama et al., 1992) are known to modulate the N400 component of ERP. The results obtained here and in Koyama et al. (1999) thus suggest that the activity in the left perisylvian area is associated with the N400. In a recent MEG study, Helenius et al. (1998) clearly showed that the cortex adjacent to the auditory area is related to the generation of the N400. They used a classical N400 paradigm in which the responses to a word, appropriate to the sentence context, were compared with those to an inappropriate word, and found that the left perisylvian activity was reduced by context appropriateness. The present find-

125

ing is thus consistent with results obtained in their study. Contrary to the present study, Rugg and colleagues (Rugg, 1987; Rugg et al., 1995; Rugg and Nagy, 1987) have obtained the nonword repetition effect on ERP. In their study, the ERPs to repeated nonwords were more positive-going than those to the first presentation of the nonwords whether they were repeated immediately (Rugg, 1987; Rugg and Nagy, 1987) or six items later (Rugg et al., 1995). One explanation for this discrepancy might be that the nature of nonwords differed between studies. The nonwords used in Rugg and colleagues were those orthographically similar to real English words (e.g., FLEEB). Their nonwords thus might partially activate the representations corresponding to the original words (FLEET). This activation might make the representation easily accessible in the second presentations, leading to the modulation of the ERP amplitude. On the other hand, our Katakana nonwords did not resemble any words although they were pronounceable (such nonwords are possible because Katakana is syllabic script in which each letter represents a single pronunciation, in contrast to English in which pronunciations are derived from legal combination of letters). Thus, in the present study, there may exist no representation that might be activated by nonwords sufficiently to induce the repetition effect. This view is consistent with the finding in Rugg and Nagy (1987) showing that the repetition of orthographically illegal nonwords (e.g., SKHRA) did not modulate the amplitude of ERPs. Alternatively, the ERP repetition effect for nonwords might reflect the modulation of activity that MEG could not detect in the present study. Magnetic fields decay rapidly as a function of distance from sensors. The fields generated in the deep brain area are thus difficult to record with sensors at the scalp. Intracranial recording studies have shown that the field potentials from the medial temporal lobe (MTL) near the hippocampus and amygdala are modulated by word repetition (Heit et al., 1990; Smith et al., 1986). This area has been also found to be related to the generation of the N400 (McCarthy et al., 1995; Nobre and McCarthy, 1995). Because the MTL is located deep in the brain, it is probable that the modulation of the MTL activity caused by nonword repetition was reflected on EEG but not on MEG. This view can be examined by recording ERP simultaneously with MEG. Representations Modulated by Word Repetition Neuropsychological studies have suggested that the orthographical, phonological, and semantic representations of words are stored in separate areas in the brain (e.g., Morton and Patterson, 1980). It is thus important to specify the type of representation stored in the left perisylvian area. Studies using positron

126

SEKIGUCHI, KOYAMA, AND KAKIGI

emission tomography (PET) and functional magnetic resonance imaging (fMRI) have reported that the left perisylvian area is activated when subjects perform phonological tasks for visual words or letters (e.g., rhyme judgment) (Fujimaki et al., 1999; Sergent et al., 1992; Paulesu et al., 1993). This area has been also found to be activated by speech stimuli (Demonet et al., 1992; Fiez et al., 1995; Petersen et al., 1989). Furthermore, recent behavioral studies have shown that visual words are recognized phonologically via a process of phonological translation of orthographic information (e.g., Lesch and Pollatsek, 1993). The stimuli used in the present study being written in syllabic Katakana allow for easy phonological translation. On these grounds, it is likely that the left perisylvian activity observed here reflects access to phonological representation via a phonologically translated code. The word repetition and semantic priming effects on the left perisylvian activity might reflect the modification of the phonological representation due to its prior activation or spreading activation from semantically related words. In the present study, the effect of word repetition was also found in the 1 M component (200 –300 ms) recorded over the both hemispheres, and it was not modulated by nonword repetition. This result suggests that the responses in the 1 M component are also related to the lexical memory process. Although we could not determine the location of source activity for this component, both MEG (e.g., Koyama et al., 1998; Salmelin et al., 1996) and intracranial recording studies (Nobre et al., 1994) have shown that the inferior occipitotemporal area was activated by visual words within the latency of 100 –300 ms poststimulus. Furthermore, recent PET studies reported that the activity in this area was reduced by word repetition (e.g., Buckner et al., 1995; Stowe et al., 1999). The source of the 1 M component thus is probably located in the bilateral inferior occipitotemporal area. This area has been assumed to be related to visual and/or orthographical analysis of words (e.g., Petersen et al., 1989, 1990). We thus conjecture that the 1 M component is related to the access to orthographic representations of words. Mechanisms for the Reduction of the Perisylvian Activity With regard to the mechanisms associated with the reduction of the perisylvian activity, we propose the following explanation. In current models of word recognition, it is assumed that a number of word representations are simultaneously activated as candidates in the early stage of word recognition (Forster, 1976; McClelland and Rumelhart, 1981; Morton, 1979). That is, when the reader encounters a word, in addition to the representation of the presented word, other representations also receive some degree of activation ac-

cording to their perceptual similarity to the stimulus. To select an appropriate candidate among them, the interactive activation model (McClelland and Rumelhart, 1981) that has dominated recent discussions assumes lateral inhibition among the representations. The reduction of the left perisylvian activity would be explained by this lateral inhibition as follows. In the second presentation, since the activation occurring in the first presentation was sustained for some time, the representation (probably phonological representation) of the stimulus word would easily gain a high level of activation sufficient to inhibit the other representations. In this situation, the other representations would be suppressed before they gain a high activation, and accordingly total amount of activation among the candidate representations became smaller than that in the first presentation. Our conjecture is that this decrease of the redundant activations is associated with the reduction of the perisylvian activity in the present study and the shorter reaction times in behavioral studies. In fact, the decrease of the redundant activation by stimulus repetition is found in the studies recording actual neural responses. Miller and Desimone (1994) reported that some neurons in the monkey’s inferior temporal lobe had a reduced response to the repeated presentation of pictures. Desimone (1996) argued that these suppressed neurons were those encoding features not needed to identify the presented picture, and the neural activity might be sharpened by stimulus repetition so that only the neurons critical to identify them had a response. In the present study, the nonwords also activated the left perisylvian area. This result appears inconsistent with the view that this area is the locus of lexical access. The nonwords used here were dissimilar to any real words and thus assumed to activate no word representation. One explanation for this contradiction might be as follows. Although the nonwords used in the present study did not resemble any words, they shared one or two syllables with some real words. It is thus probable that our nonwords slightly activated a number of word representations. In our conjecture, the left perisylvian activity to the nonwords might derive from a large number of weakly activated representations, whereas that to the words might derive from a smaller number of strongly activated representations. The finding consistent with this notion was reported in an invasive recording study. Rainer and Miller (2000) reported that a number of neurons in the monkey’s prefrontal cortex responded to visual objects completely unfamiliar to them, and the number of responding neurons was larger to unfamiliar objects than that to familiar object. The present study revealed that the 2 M component for the nonwords was sustained for a longer duration than that for the words. This result can be also explained by the concept of multiple activation and lat-

WORD REPETITION EFFECT ON MEG

eral inhibition. In word processing, the total activation in lexical memory would soon decrease because stimulus representation suppressed the other representations. In contrast, in nonword processing, since no single representation was dominant, the representations activated by nonwords might be kept active for a while, leading to the longer activity in the brain. In sum, the results obtained here can be explained by the assumptions proposed in the interactive activation model. Conclusion We investigated neural correlates of the word repetition effect using MEG. The activity in the left perisylvian area adjacent to the auditory cortex was reduced by word repetition. It is suggested that the left perisylvian area is one locus of the word repetition effect, the activity of which might be related to the N400 component of scalp recorded ERP. The nonword repetition did not modulate the perisylvian activity. This result indicates that the left perisylvian area is involved in the lexical memory process. We propose that the activity in this area reflects simultaneous activation of a large number of word representations, and its reduction might be associated with the decrease of the redundant activation. ACKNOWLEDGMENTS We thank Professor M. D. Rugg for his valuable advice. We also thank P. Ferrari for his comments on an earlier version of the article, O. Nagata and Y. Takeshima for technical assistance, and Dr. S. Miyauchi for allowing us to utilize MRI facilities. This study was supported by a Grant-in-Aid for Scientific Research, Ministry of Education, Science, Sports and Culture, Japan, to Sachiko Koyama (No. 09710064) and Ryusuke Kakigi (No. 08878160, 09558102).

REFERENCES Bentin, S., McCarthy, G., and Wood, C. C. 1985. Event-related potentials, lexical decision and semantic priming. Electroencephalogr. Clin. Neurophysiol. 60: 343–355. Bentin, S., and Peled, B. S. 1990. The contribution of task-related factors to ERP repetition effects at short and long lags. Mem. Cognit. 18: 359 –366. Buckner, R. L., Petersen, S. E., Ojemann, J. G., Miezin, F. M., Squire, L. R., and Raichle, M. E. 1995. Functional anatomical studies of explicit and implicit memory retrieval tasks. J. Neurosci. 15: 12–29. Demonet, J. F., Chollet, F., Ramsay, S., et al. 1992. The anatomy of phonological and semantic processing in normal subjects. Brain 115: 1753–1768. Desimone, R. 1996. Neural mechanisms for visual memory and their role in attention. Proc. Natl. Acad. Sci. USA 93: 13494 –13499. Durso, F. T., and Johnson, M. K. 1979. Facilitation in naming and categorizing repeated pictures and words. J. Exp. Psychol. Hum. Mem. Cogn. 5: 449 – 459. Feustel, T. C., Shiffrin, R. M., and Salasoo, A. 1983. Episodic and lexical contributions to the repetition effect in word identification. J. Exp. Psychol. Gen. 112: 309 –346.

127

Fiez, J. A., Raichle, M. E., Miezin, F. M., Petersen, S. E., Tallal, P., and Katz, W. F. 1995. PET studies of auditory and phonological processing: Effects of stimulus characteristics and task demands. J. Cogn. Neurosci. 7: 357–375. Forster, K. I. 1976. Accessing the mental lexicon. In New Approaches to Language Mechanisms (R. J. Wales and E. Walker, Eds.), pp. 257–288. North-Holland Press, Amsterdam. Fujimaki, N., Miyauchi, S., Putz, B., et al. 1999. Functional magnetic resonance imaging of neural activity related to orthographic, phonological, and lexico-semantic judgments of visually presented characters and words. Hum. Brain Mapp. 8: 44 –59. Ha¨ma¨la¨inen, M., Hari, R., Ilmoniemi, R. J., Knuutila, J., and Lounasmaa, O. V. 1993. Magnetoencephalography—Theory, instrumentation, and applications to noninvasive studies of the working human brain. Rev. Mod. Phys. 65: 413– 497. Hayashi, T. 1976. Nonsense Syllable Shin-kijunhyo (A New Norm of Japanese Nonsense Syllables). Tokai Univ. Press, Tokyo. Heit, G., Smith, M. E., and Halgren, E. 1990. Neuronal activity in the human medial temporal lobe during recognition memory. Brain 113: 1093–1112. Helenius, P., Salmelin, R., Service, E., and Connolly, J. F. 1998. Distinct time courses of word and context comprehension in the left temporal cortex. Brain 121: 1133–1142. Jacoby, L. L. 1983. Perceptual enhancement: Persistent effects of an experience. J. Exp. Psychol. Learn. Mem. Cogn. 9: 21–38. Kaneoke, Y., Bundou, M., Koyama, S., Suzuki, H., and Kakigi, R. 1997. Human cortical area responding to stimuli in apparent motion. Neuroreport 8: 677– 682. Koyama, S., Gunji, A., Yabe, H., et al. 2000. The masking effect in foreign speech sounds perception revealed by neuromagnetic responses. Neuroreport 11: 3765–3769. Koyama, S., Kakigi, R., Hoshiyama, M., and Kitamura, Y. 1998. Reading of Japanese Kanji (morphograms) and Kana (syllabograms): A magnetoencephalographic study. Neuropsychologia 36: 83–98. Koyama, S., Nageishi, Y., and Shimokochi, M. 1992. Effects of semantic context and event-related potentials: N400 correlates with inhibition effect. Brain Lang. 43: 668 – 681. Koyama, S., Naka, D., and Kakigi, R. 1999. Evoked magnetic responses during a word completion task. Electroencephalogr. Clin. Neurophysiol. Suppl. 49: 174 –178. Lesch, M. F., and Pollatsek, A. 1993. Automatic access of semantic information by phonological codes in visual word recognition. J. Exp. Psychol. Learn. Mem. Cogn. 19: 285–294. McCarthy, G., Nobre, A. C., Bentin, S., and Spencer, D. D. 1995. Language-related field potentials in the anterior-medial temporal lobe: I. Intracranial distribution and neural generators. J. Neurosci. 15: 1080 –1089. McClelland, J. L., and Rumelhart, D. E. 1981. An interactive activation model of context effects in letter perception: I. An account of basic findings. Psychol. Rev. 88: 375– 407. Miller, E. K., and Desimone, R. 1994. Parallel neuronal mechanisms for short-term memory. Science 263: 520 –522. Monsell, S. 1985. Repetition and the lexicon. In Progress in the Psychology of Language (A. W. Ellis, Ed.), Vol. 2, pp. 147–195. Lawrence Erlbaum, London. Morton, J. 1979. Facilitation in word recognition: Experiments causing change in the logogen model. In Processing of Visible Language (P. A. Kolers, M. E. Wrolstad, and H. Bouma, Eds.), Vol. 1, pp. 259 –268. Plenum Press, New York. Morton, J., and Patterson, K. 1980. A new attempt at an interpretation, or an attempt at a new interpretation. In Deep Dyslexia (M. Coltheart, K. Patterson, and J. C. Marshall, Eds.), pp. 91–118. Routledge & Kegan Paul, London.

128

SEKIGUCHI, KOYAMA, AND KAKIGI

Nagy, M. E., and Rugg, M. D. 1989. Modulation of event-related potentials by word repetition: The effects of inter-item lag. Psychophysiology 26: 431– 436. Neely, J. H. 1991. Semantic priming effects in visual word recognition: A selective review of current findings and theories. In Basic Processes in Reading: Visual Word Recognition (D. Besner and G. W. Humphreys, Eds.), pp. 264 –336. Lawrence Erlbaum, Hillsdale, NJ. Nobre, A. C., Allison, T., and McCarthy, G. 1994. Word recognition in the human inferior temporal lobe. Nature 372: 260 –263. Nobre, A. C., and McCarthy, G. 1995. Language-related field potentials in the anterior-medial temporal lobe: II. Effects of word type and semantic priming. J. Neurosci. 15: 1090 –1098. Paulesu, E., Frith, C. D., and Frackowiak, R. S. J. 1993. The neural correlates of the verbal component of working memory. Nature 362: 342–345. Petersen, S. E., Fox, P. T., Posner, M. I., Mintun, M., and Raichle, M. E. 1989. Positron emission tomographic studies of the processing of single words. J. Cogn. Neurosci. 1: 153–170. Petersen, S. E., Fox, P. T., Snyder, A. Z., and Raichle, M. E. 1990. Activation of extrastriate and frontal cortical areas by visual words and word-like stimuli. Science 249: 1041–1044. Rainer, G., and Miller, E. K. 2000. Effects of visual experience on the representation of objects in the prefrontal cortex. Neuron 27: 179 – 189. Rugg, M. D. 1987. Dissociation of semantic priming, word and nonword repetition effects by event-related potentials. Q. J. Exp. Psychol. A39: 123–148. Rugg, M. D. 1990. Event-related brain potentials dissociate repetition effects of high- and low-frequency words. Mem. Cogn. 18: 367–379. Rugg, M. D., and Doyle, M. C. 1994. Event-related potentials and stimulus repetition in direct and indirect tests of memory. In Cognitive Electrophysiology (H. Heinze, T. Munte, and G. R. Mangun, Eds.), pp. 124 –148. Birkhauser, Boston, MA. Rugg, M. D., Doyle, M. C., and Wells, T. 1995. Word and nonword repetition within- and across-modality: An event-related potential study. J. Cogn. Neurosci. 7: 209 –227. Rugg, M. D., and Nagy, M. E. 1987. Lexical contribution to nonwordrepetition effects: Evidence from event-related potentials. Mem. Cogn. 15: 473– 481.

Rugg, M. D., Schloerscheidt, A. M., Doyle, M. C., Cox, C. J. C., and Patching, G. R. 1996. Event-related potentials and the recollection of associative information. Brain Res. Cogn. Brain Res. 4: 297–304. Salasoo, A., Shiffrin, R. M., and Feustel, T. C. 1985. Building permanent memory codes: Codification and repetition effects in word identification. J. Exp. Psychol. Gen. 114: 50 –77. Salmelin, R., Service, E., Kiesila¨, P., Uutela, K., and Salonen, O. 1996. Impaired visual word processing in dyslexia revealed with magnetoencephalography. Ann. Neurol. 40: 157–162. Sarvas, J. 1987. Basic mathematical and electromagnetic concepts of the biomagnetic inverse problem. Phys. Med. Biol. 32: 11–22. Scarborough, D. L., Cortese, C., and Scarborough, H. S. 1977. Frequency and repetition effects in lexical memory. J. Exp. Psychol. Hum. Percept. Perform. 3: 1–17. Scherg, M. 1992. Functional imaging and localization of electromagnetic brain activity. Brain Topogr. 5: 103–111. Sekiguchi, T., Koyama, S., and Kakigi, R. 2000. The effect of word repetition on evoked magnetic responses of the human brain. Jpn. Psychol. Res. 42: 3–14. Sergent, J., Zuck, E., Levesque, M., and MacDonald, B. 1992. Positron emission tomography study of letter and object processing: Empirical findings and methodological considerations. Cereb. Cortex 2: 68 – 80. Smith, M. E., Stapleton, J. M., and Halgren, E. 1986. Human medial temporal lobe potentials evoked in memory and language tasks. Electroencephalogr. Clin. Neurophysiol. 63: 145–159. Snodgrass, J. G., and Corwin, J. 1988. Pragmatics of measuring recognition memory: Applications to dementia and amnesia. J. Exp. Psychol. Gen. 117: 34 –50. Stowe, L. A., Paans, A. M. J., Wijers, A. A., et al. 1999. Sentence comprehension and word repetition: A positron emission tomography investigation. Psychophysiology 36: 786 – 801. Tenpenny, P. L. 1995. Abstractionist versus episodic theories of repetition priming and word identification. Pshchono. Bull. Rev. 2: 339 –363. Van Petten, C., Kutas, M., Kluender, R., Mitchiner, M., and McIsaac, H. 1991. Fractionating the word repetition effect with event-related potentials. J. Cogn. Neurosci. 3: 131–150.