Processes involved in tempo perception: A CNV analysis

Transcription

1 Psychophysiology, 40 (2003), Blackwell Publishing Inc. Printed in the USA. Copyright r 2003 Society for Psychophysiological Research Processes involved in tempo perception: A CNV analysis MICHA PFEUTY, RICHARDRAGOT, AND VIVIANE POUTHAS Neurosciences Cognitives et Imagerie Ce re brale, LENA-CNRS, Hoˆ pital Salpêtrie re, Paris Cedex 13, France Abstract The purpose of this research was to study the mechanisms underlying tempo perception, by looking at their electrophysiological brain correlates. The subjects task consisted of comparing the tempos of two isochronous tone sequences made up of either three (condition I3) or six (condition I6) 600-ms intervals. Contingent negative variation (CNV), known to be linked to the judgment of a single interval, kept increasing in amplitude for three intervals during tempo encoding, thereby providing evidence of the occurrence of CNVs also for several intervals in succession. This CNV increase could reflect the use of interval-based processes in the building of the interval memory trace. During the comparison phase, a CNV decrease was observed in condition I6, suggesting that subjects did not build a new memory trace, but used beat-based processes to check whether the beats of the new tempo occurred at the times they anticipated. Descriptors: Time perception, Tempo, EEG, CNV, Memory, Audition Many studies have been devoted to the analysis of the processes underlying time perception (for reviews, see Gibbon, Malapani, Dale, & Gallistel, 1997; Harrington & Haaland, 1999; Ivry, 1996; among others). Several questions remain unanswered, and an interesting way to approach this issue is to study the mechanisms that subserve the perception of a series of time intervals, especially in the auditory modality. We all experience such perceptions daily, particularly in language and music. This article will focus on a basic aspect of rhythm perception, tempo, defined as a sequence of tones separated by isochronous intervals. Models of time perception generally assume that the duration of an interval is measured by an internal timer that records the number of clock ticks occurring during the interval. The judgment that a test interval is the same as, shorter than, or longer than a target interval would result from the comparison of the number of ticks recorded during each interval. The accuracy of such a comparison would depend on the variability of the timer. It must be stressed that few studies have taken an interest in the effect of the temporal context (e.g., sequences of intervals, as in speech or music) on judgment of interval duration. However, two types of modelsfinterval-based and beat-based modelsfhave been proposed to account for timing data in these particular contexts. According to interval-based models, the comparison of intervals relies on the timer described above, which is thought to be involved in the estimation of successive intervals. According to beat-based models, such a comparison relies on the synchronization of an internal beat with the external rhythm. Address reprint requests to: Micha Pfeuty, Neurosciences Cognitives et Imagerie Cérébrale, LENA-CNRS, Hoˆ pital Salpêtrière, 47 Bld de l Hoˆ pital, Paris Cedex 13, France. Micha.Pfeuty@chups.- jussieu.fr. 69 Schulze (1978) conducted an experiment in which subjects had to decide whether an auditory sequence was irregular. The author measured the threshold, dt, for the detection of an irregularity in a sequence of type A (T, T, T, T 1 dt, T, T, T, y) or B (T, T, T, T 1 dt, T dt, T, T, y). If performance is subserved by beatbased processes, a lower threshold for detecting irregularity will be observed in sequence A, because the difference between internal and external rhythm would be maintained throughout the sequence. The involvement of interval-based processes would give a lower threshold in sequence B, because the differences T1dT and T dt would be more evident for the timer. The results showed that the threshold for the detection of irregularity was lower in sequence A than in sequence B, as predicted by beatbased models. Oddly, Keele, Nicholetti, Ivry, and Pokorny (1989), who replicated the same experiment, reported the opposite results, and thus proposed that interval-based models better account for performance. However, this last experiment was not sufficient for choosing between the two kinds of models. So in another experiment, in which subjects had to compare the tempos of two isochronous tone sequences, Keele et al. manipulated the duration of the pause between the target sequence and the test sequence. They assumed that, if timing was based on an internal beat, varying the duration of this pause would disturb the comparison. The results did not indicate any significant effect of the pause, thereby refuting beat-based models. Ivry and Hazeltine (1995) examined time discrimination for different interval durations in order to describe the durationdependent variability of the timer. They used either a single interval or a sequence of three intervals. The results showed that the duration-dependent variability of the timer was smaller in the second condition, suggesting that repeating the target interval allows participants to form a more accurate internal representation. In a similar way, Drake and Botte (1993) showed that the threshold of tempo discrimination decreased when the number of

2 70 M. Pfeuty, R. Ragot, and V. Pouthas intervals in the sequence increased. They proposed a multiplelook model of tempo perception. According to these authors, when a standard interval is repeated, multiple looks at the same interval would allow listeners to improve the statistical estimate of it. Thus, interval-based models account for context effects in terms of the statistics of the intervals constituting a pattern. Tempo encoding would rest on the elaboration of a memory trace of the reference interval equal to the statistical mean of the successive intervals. This trace is thought to become more precise as the number of intervals included in the tempo increases. Nevertheless, Panissal (1998) showed that the threshold of tempo discrimination ceases to decrease after a critical number of intervals, depending on the reference interval duration. This suggests the existence of a temporal window, which would be equal to the minimal number of intervals needed to establish the most accurate internal interval reference. This temporal window is then said to be filled after that number of intervals. In short, there have been behavioral studies on tempo perception guided by either interval-based or beat-based models (Drake & Botte, 1993; Ivry & Hazeltine, 1995; Keele et al., 1989; Schulze, 1978; among others), but to our knowledge, there have been no EEG studies like those conducted on the perception and the production of a single interval. A slow negative wave, called the contingent negative variation (CNV), has been shown to develop over frontal areas during the temporal interval between two events, S1 corresponding to a warning stimulus, and S2 to an imperative stimulus (Walter, Cooper, Aldridge, McCallum, & Winter, 1964). This wave would reflect anticipatory processing of S2 and motor preparation when a response is temporally related to S2. Several lines of evidence support the hypothesis that the CNV is the main ERP correlate of the estimation or production of a time interval. McAdam s (1966) results showed that CNV amplitude and accuracy increased in parallel during the learning of a temporal interval. Ruchkin, McCalley, and Glaser (1977) observed that the latency of CNV termination was correlated with temporal judgments and concluded that this ERP partly reflected cognitive activity leading up to the formation of a temporal judgment. In Macar, Vidal, and Casini (1999), larger negativities were found either when longer intervals than the target were produced or when intervals were judged longer than a memorized standard interval. They proposed that these negativities would reflect the accumulation of pulses during the interval to be estimated. The aim of the present study was to examine the ERP correlates of the estimation of intervals that form an isochronous sequence, and to find out if and how successive CNVs are related to tempo perception. Based on behavioral results on a tempo comparison task (Drake & Botte, 1993; Panissal, 1998) and on the assumption that CNV reflects the operation of a timer, we hypothesized the following: If tempo encoding relies on the building of a memory trace issued from the multiple estimation of successive intervals, EEG recordings will reveal a systematic increase in CNV amplitude reflecting the temporal judgment for those intervals. We also assumed that intervals going beyond a definite temporal window would no longer be taken into account, and that CNV amplitude would therefore stop increasing. For a tempo period of 600 ms, behavioral studies have suggested that this temporal window is filled after three or four intervals (Panissal, 1998). In the present study, subjects were presented with a tempo made up of either three (condition I3) or six (condition I6) intervals. We expected to observe a systematic increase in CNV amplitude throughout the encoding phase in condition I3, and no further increase after three or four intervals in condition I6. A complementary goal of the present study was to examine ERPs during the comparison phase. Drake and Botte (1993) proposed that subjects would directly compare the successive intervals with the average memory trace previously elaborated. According to this assumption, no sustained negativity should then be observed during the comparison phase. Method Participants Fifteen volunteer students (8 men, 7 women) ages participated in this experiment. They reported being righthanded, having normal hearing, and having never attended music school or played a musical instrument. Procedure Tempos were made of regular sequences of auditory tones presented at a regular interval. Each tone was made of a 440-Hz sine wave, 50 ms in duration, presented binaurally at 70 db (SPL). The tempo was defined by the interonset interval (IOI) between two successive tones. Each trial consisted of two sequences (standard and test sequences) separated by a random delay (between 1,100 and 1,500 ms long). The IOI defining the tempo of the first sequence was always 600 ms long (standard IOI). The test IOI defining the tempo of the second sequence was either the same (600 ms 5 standard), shorter (576 ms ms 4% of 600 ms), or longer (624 ms ms þ 4% of 600 ms; Figure 1). The participants task consisted in comparing the tempo of the second test sequence with that of the first standard sequence. They were given a response device with three keys, and had to respond using the three inner fingers of their right hand. The sequences consisted of either three intervals (condition I3) or six intervals (condition I6). The order of conditions I3 and I6 was counterbalanced among the participants. For each condition, there were 96 trials, each divided into four sections of 24. The total number of trials was the same (1/3 of ) for each test IOI category (standard, short, or long). Participants used one of the three keys to state whether they judged the second tempo to be similar (standard IOI, middle finger), faster (shorter IOI, index finger), or slower (longer IOI, ring finger) than the first tempo. They were asked to respond as accurately as possible, without worrying about speed. They had to wait for the end of the second tempo sample before responding, even if they had already decided which button to press. ERP Recording Participants sat comfortably in a dimly lit room. During the task, they were asked to focus their gaze on a red point located 1 m in front of them. EEG activity was recorded through 62 electrodes placed at standard electrode sites and referred to the earlobes. The sampling rate per channel was 500 Hz (bandwidth Hz). Horizontal and vertical eye movements and blinks (EOG) were recorded by two bipolar electrodes. EEG signals were automatically corrected for ocular artifacts using an off-line digital method (Gratton, Coles, & Donchin, 1983). Eventrelated potentials were obtained by averaging trials in each condition (I3, I6), phase (encoding, comparison), and test IOI length (short, standard, long). There was a 200-ms baseline

3 Processes involved in tempo perception 71 Figure 1. Diagram of the tempo encoding and comparison paradigm. At the end of the comparison phase, subjects had to decide whether the presented tempo was slower than, equal to, or longer than the standard tempo presented before in the encoding phase. In condition I3, because of the small number of intervals (three) in the encoding and comparison phases, the temporal window (see text) can be considered as hardly filled. In condition I6, because of the large number of intervals (six) in the encoding and comparison phases, the temporal window can be considered as completely filled. epoch prior to the first tone of each sequence. This was used as a reference for amplitude evaluation. Behavioral Analysis Two performance indexes were used during this experiment: (1) The d 0 measure of detection sensitivity from signal detection theory, known to be an indicator of judgment difficulty independent of the strategy adopted by the subject (bias to respond different or identical ), was measured for each subject in conditions I3 and I6. The correct detection rate was obtained by averaging the percentages of correct responses for short and long test IOIs, and the false alarm rate was equal to the percentage of false responses for the standard IOI. The d 0 measure was calculated using the following formula: d 0 ¼ z fa z cd ; where z fa is the false alarm rate and z cd is the correct detection rate, both converted into z units. (2) The mean response latency, measured from the beginning of the last tone of the sequence, was calculated for each test IOI length and for each participant. The response latencies for short and long test IOIs were averaged and gave the correct detection latency, and the latency on the standard test IOI was the correct rejection latency. The effect of the condition (I3, I6) on d 0 was tested using a one-way ANOVA. The effect of the condition (I3, I6) and the type of response (detection, rejection) on the response latency was tested using a two-way ANOVA (Condition Type of Response). ERP Analysis CNVamplitude: Grand mean analysis. Mean CNVamplitude was measured over frontal electrodes (F5, F3, F1, FZ, F2, F4, F6, FC5, FC3, FC1, FCZ, FC2, FC4, FC6) for each interval, from 250 ms after each tone until the next tone. It was then averaged over all frontal sites in conditions I3 and I6 during the encoding and comparison phases. ANOVAS were carried out on these measures, to test whether or not there was an increase in CNV amplitude during these two phases. It was subsequently averaged for three groups of frontal sites, left (F5, F3, FC5, FC3), medial (F1, FZ, F2, FC1, FCZ, FC2), and right (F4, F6, FC4, FC6), to determine whether or not a specific hemispheric distribution of CNV was observed. Greenhouse Geisser corrections were made to account for the fact that the measurements of CNV amplitude on successive intervals were not independent. CNVamplitude: Individual analysis. In addition to the grand mean analysis, individual analyses were performed on CNV amplitude during the encoding phase, at the FCZ electrode where the CNV peaked. Given that we were interested in the time-course analysis of CNV amplitude (i.e., increase or decrease), individual differences in the overall CNV amplitude were eliminated. To do this, the amplitude at FCZ was replaced by a normalized z value for each subject: z ¼ ða FCS m ½b:eŠ Þ=s ½b:eŠ ; where A FCZ is the amplitude at FCZ, and m [b:e] and s [b:e] are the average and standard deviation, respectively, of the successive amplitudes measured at FCZ between the beginning and the end of the encoding sequence (from 0 to 1,800 ms in condition I3 and from 0 to 3,600 ms in condition I6). (Note: the minus sign in front of the formula allows an increase in the z value to correspond to an increase in CNV amplitude.) To characterize the time course of the z value along the encoding sequence, a mean z value was calculated during each interval, from 250 ms after each tone until the next tone. Then a z-slope value was obtained by linear regression of the different mean z values measured over the successive intervals (between the first and the last). This z-slope value was expressed in z units per interval. A positive and a negative z-slope value meant an increase and a decrease in CNV amplitude, respectively. Results Behavioral Data The d 0 measure of detection sensitivity was significantly greater in condition I6 than in condition I3, F(1,14) , po.05. Tempo discrimination accuracy improved between condition I3, d , and condition I6, d

4 72 M. Pfeuty, R. Ragot, and V. Pouthas The ANOVA on response latency indicated an effect of the condition, F(1,14) , po.01, and of the type of response, F(1,14) , po.005. There was no interaction between these two factors, F(1,14) , p4.10. The response latency was shorter in condition I6 (837 ms) than in condition I3 (956 ms), and shorter for correct detection (796 ms) than for correct rejection (1,097 ms). ERP Data CNV amplitude: Grand mean analysis. Figure 2 presents the grand mean ERPs over frontal electrodes in conditions I3 and I6 for the encoding and comparison phases. Mean CNVamplitudes measured for each interval, from 250 ms after each tone until the next tone, were averaged over frontal electrodes to obtain the mean frontal amplitude. To find out whether there was an increase in the mean frontal amplitude during the encoding and comparison phases, the effect of the interval rank (R1 to R3 in condition I3 and R1 to R6 in condition I6) was tested using a one-way ANOVA. For the encoding phase, the ANOVA yielded an effect of the interval rank in condition I3, F(2,28) , po.05, e 5.52, and in condition I6, F(5,70) , p o.05, e Table 1, part A, shows the mean CNV amplitude for successive intervals in both conditions. To test the hypothesis that a systematic increase in CNV amplitude would be observed throughout the encoding phase in condition I3, and no further increase after three or four intervals in condition I6, planned comparisons were carried out between successive interval ranks. In condition I3, the negativity increased between R1 and R2, F(1,14) , po.05, and tended to increase between R2 and R3, F(1,14) , po.10. In condition I6, the negativity increased between R1 and R2, F(1,14) , po.05, and between R2 and R3, F(1,14) , po.001, but not between R3 and R4, F(1,14) , p4.10, between R4 and R5, F(1,14) , p4.10, or between R5 and R6, F(1,14) , p4.10. Thus, during the encoding phase, an increase in CNV amplitude was only observed up to the third interval. For the comparison phase, there was no effect of the test IOI length factor in condition I3, F(2,28)o1, n.s., or I6, F(2,28) , p4.10, so this factor was removed from subsequent analyses. The interval rank effect was tested using a one-way ANOVA. No effect was observed on CNVamplitude in condition I3, F(2,28) , p4.10, whereas an effect was observed in condition I6, F(5,70) , po.0005, e Table 1, Part B, shows the mean CNV amplitude for successive intervals in both conditions. As no a priori assumptions have Figure 2. Grand mean event-related potentials (ERPs) over frontal (F5, F3, F1, FZ, F2, F4, F6) and frontal-central electrodes (FC5, FC3, FC1, FCZ, FC2, FC4, FC6), elicited during tempo encoding (thick lines) and tempo comparison (thin lines) in the six-interval and three-interval conditions. been made, post hoc comparisons using Tukey s HSDmeasures were carried out. These showed that in condition I6, the negative potential did not vary between R1 and R2, then decreased and changed into a positive potential between R2 and R4, po.001. Thus, CNVamplitude was sustained in condition I3, whereas in condition I6, it decreased after the second interval. To determine whether or not a hemispheric-specific distribution of the CNV was observed over the scalp, the left-frontal amplitude resulting from the averaging of electrodes F5, F3, Table 1. Mean CNV Amplitude Measured over Frontal Electrodes for Successive Intervals in Both Conditions I3 and I6 for Both Encoding (A) and Comparison (B) Conditions Interval rank Condition I3 Condition I6 R1 R2 R3 R1 R2 R3 R4 R5 R6 Part A: Encoding phase CNV amplitude (mv) Part B: Comparison phase CNV amplitude (mv)

5 Processes involved in tempo perception 73 FC5, and FC3 was compared to the right-frontal amplitude resulting from the averaging of electrodes F4, F6, FC4, and FC6. The laterality effect was tested using a two-way ANOVA (Interval Rank Laterality) in both the encoding and comparison phases. No main effect of laterality was observed on the encoding and comparison phases in either condition I3 or I6. An interaction was observed for the encoding phase in condition I3 only, F(2,28) , po.05. Post hoc comparisons showed that during the first interval, left ( 4.17 mv) and right ( 3.82 mv), p4.10, frontal negativity did not differ significantly whereas during the third interval, frontal negativity was significantly larger on the left ( 7.51 mv) than on the right ( 6.53 mv), p o.001. CNV amplitude: Individual analysis. Figure 3 presents the time course of the individual ERPs at electrode FCZ, first transformed into a normalized z value, in both conditions (I3 and I6). Individual z-slope values (in z units per interval) are displayed below. The individual data showed that for 11 out of 15 subjects, the z-slope value, which is equivalent to the CNV amplitude slope at the FCZ electrode, was positive in at least one of the two conditions. By contrast, for 4 subjects out of 15, the slope was negative in both conditions. On the basis of these results, subjects were separated into two groups: group G1 consisting of 11 subjects, whose overall slope was positive, and group G2 consisting of 4 subjects whose overall slope was negative. For G1 subjects, CNVamplitude increased throughout the encoding phase, whereas for G2 subjects, CNV amplitude was high during the first interval but decreased after that. This is illustrated in Figure 4, which presents the time course of CNV amplitude over frontal electrodes, averaged separately for groups G1 and G2. To find out whether this group difference occurred over all frontal electrodes, frontal activity was averaged separately for three groups of frontal sites: left (F5, F3, FC5, FC3), medial (F1, FZ, F2, FC1, FCZ, FC2), and right (F4, F6, FC4, FC6). A three-way ANOVA was performed (Group Interval Rank Location). The results indicated a Group Interval Rank interaction effect in conditions I3, F(2,26) , po.05, e 5.53, and I6, F(5,65) , po.01, e 5.29, showing that in both conditions, CNV amplitude increased throughout the sequence for G1 subjects while decreasing after the end of the first interval for G2 subjects. Moreover, no interaction effect was observed with the location factor. Thus, the difference in the CNV amplitude slope between the two groups did not vary significantly across the different frontal sites. We therefore checked whether these different CNV patterns corresponded to different behaviors. The results showed that the behavioral performance of G2 subjects was significantly better than that of G1 subjects (mean d vs. 1.04, po.01, Mann Whitney U test). This finding is an additional argument for separating subjects into two groups. Discussion Behavioral Data The behavioral results showed that the tempo comparison task was easier with six than with three intervals. The d 0 measure of detection sensitivity rose, and response latency fell between conditions I3 and I6. This suggests (a) that three intervals are not sufficient for achieving optimal performance in the comparison of two tempos with a 600-ms interval, and (b) that temporal information after the third interval is still useful in enhancing performance. Panissal s (1998) data indicated that three or four intervals might be sufficient to encode a tempo of 600 ms. In our experiment, though, subjects did not undergo any special training, unlike most psychophysical studies. This may be why they needed more intervals to optimize their performance. However, it is not possible to determine whether the increase in sensitivity observed in condition I6 was due to supplementary intervals during the encoding or the comparison phase. Response latency was greater for correct rejection (equal tempo) than for correct detection (slower or longer tempo). This result suggests that the relative judgment of two tempos is not based on recognition processes, but on the detection of differences. ERP Data CNV Amplitude: Grand Mean Analysis Encoding phase. The grand mean ERPs over frontal electrodes showed that in conditions I3 and I6 alike, CNV amplitude increased during the encoding phase over the frontal part of the scalp (Figure 2). Such an increase did not occur during the comparison phase, and a decrease was even observed in condition I6. Thus, the increase in CNV amplitude seems to reflect processes strongly related to tempo encoding. Several studies have shown the CNVto be linked to time processing for a single interval (Macar et al., 1999; McAdam, 1966; Ruchkin et al., 1977). Our data provide evidence of CNVs also occurring for several intervals in succession. This suggests that during tempo encoding, intervals are estimated in succession, in accordance with the predictions of interval-based models of rhythm perception (Drake & Botte, 1993; Ivry & Hazeltine, 1995; Keele et al., 1989). We therefore propose that such an increase in CNV amplitude would reflect the building of a memory trace of the reference interval. Moreover, in condition I6, the results showed that the negativity stopped increasing after the third interval. Thus, beyond a critical number of intervals (defined as the temporal window), the memory trace would be optimal and time intervals would no longer be estimated. Two alternative hypotheses can be put forward to explain how the increase in CNVamplitude might be linked to the building of the interval memory trace. Regarding the first, McAdam (1966) suggested that CNV amplitude depends on the amount of attention allocated to time estimation. His study showed that CNV amplitude increased when the subjects learned a duration, and decreased after the learning phase (when performance became maximal). The author proposed that subcortical structures may take over for cortical structures when the task becomes automatic. Thus, in the present study, an increase in CNV amplitude could reflect attention allocation for successive time intervals. Once the memory trace became strong enough after three intervals, the role of attentional processes devoted to time would decrease and tempo would be perceived in a more automatic and efficient way until the end of the sequence. Regarding the second hypothesis, Ruchkin et al. (1997) showed that when subjects had to retain the presentation order of a series of syllables in working memory, a bilateral frontal slow negativity developed for syllables presented in the auditory modality but not for those presented in the visual modality. These authors proposed that this bilateral frontal slow negativity reflects the automatic storage of auditory information. Accordingly, because our temporal intervals were defined by tones, the

6 74 M. Pfeuty, R. Ragot, and V. Pouthas Figure 3. Time course of the individual ERPs at electrode FCZ (transformed into a normalized z value) elicited by tempo encoding with six (upper part) and three intervals (lower part), and grouped separately for G1 subjects (blue-green colors on left) and for G2 subjects (red colors on right). The corresponding individual z-slope values (in z units per time interval) are displayed below the curves. increase in CNVamplitude found in the present study may reflect the automatic storage of the successive tones in working memory. This type of storage could thus have been used to build the memory trace of the interval. Comparison phase. During the comparison phase, a decrease was observed in condition I6 over the frontal part of the scalp (Figure 2), after the second interval. This suggests that subjects did not build a new memory trace during this comparison phase. They may have directly compared the trace elaborated during the encoding phase with the current intervals. Decisional processes should play a more important role in order to make such a comparison. Numerous studies have shown that positive waves (P300-like) can reflect decisional processes taking place after the CNV (for a review, see Timsit-Berthier, 1984). Thus, the decrease in CNV amplitude observed on the comparison phase might

7 Processes involved in tempo perception 75 come from a decrease in attentional or working memory processes (generating less negativity), together with a possible increase in decisional processes (generating more positivity). Our hypothesis is that the subjects first mentally rehearsed the memorized tempo (internal), and then checked whether the beats of the new tempo (external) occurred at the times they anticipated. This is consistent with beat-based models (Schulze, 1978). In condition I3, the grand mean ERPs did not exhibit a decrease in CNVamplitude (Figure 2). This could result from less accurate encoding of the tempo, as suggested by the poorer d 0 obtained in this condition. The explanatory hypothesis we can propose is that some subjects encoded a new memory trace during the comparison phase in order to make their judgment. Laterality analysis. For the encoding phase in condition I3, the results showed that, over time, the left frontal hemisphere became more active than the right frontal hemisphere. Given that the laterality effect did not occur in condition I6, this result suggests that left frontal activity was enhanced when the task was more difficult (i.e., when the number of intervals was smaller). Ruchkin et al. (1997) showed, using a working memory task, that a left frontal slow wave occurred for auditory rehearsal. Because in condition I3 the reference interval was apparently not repeated enough to be thoroughly encoded, subjects may have used auditory rehearsal as a compensatory strategy in a way that reinforced their interval memory trace. Such a strategy may not have been necessary in condition I6 due to the larger number of intervals. CNV Amplitude: Individual Analysis Individual analyses of CNV slope calculated at electrode FCZ (where the grand mean CNV peaked) showed that subjects could be separated into two groups (Figure 3). For 11 subjects out of 15 (group G1), the CNV slope was globally positive (increase in CNV amplitude) in both conditions (I3 and I6) whereas for 4 subjects out of 15 (group G2), the CNV slope was negative (decrease in CNV amplitude) in both conditions. This group difference in the CNV pattern occurred over all frontal electrodes (Figure 4). The behavioral performance of G2 subjects was significantly better than that of G1 subjects, further supporting this subject division. Thus, subjects who failed to show an increase in CNVamplitude during the encoding phase performed better. This result corroborates numerous studies using temporal tasks that have found a higher CNVamplitude for poor than for good performers (Ladanyi & Dubrovsky, 1985) or greater prefrontal activity for incorrect than for correct responses (Casini & Macar, 1996). The same result was also found in a mental imagery task, where the authors showed that the left frontal slow wave amplitude was larger for poor than for good performers (Ro sler, Heil, & Ro der, 1997). How can one explain why a decrease in CNV amplitude was observed for G2 subjects during the encoding phase? G2 subjects may have an accurate estimation of the very first interval, which directly provided them with a memory trace of the reference interval. They would then only have to check if this trace was correct during subsequent intervals, that is, if it allowed for a good prediction about the temporal occurrence of subsequent tones. Thus, they would behave in the same way as during the comparison phase (Figure 4). Their memory trace could be maintained by a checking process. This interpretation is consistent with beat-based models. Such an encoding strategy would enhance performance, while being more economical by not requiring reliance on high-level cognitive functions such as working memory or attention. General Discussion The main goal of our study was to determine how the CNV develops during tempo encoding and how it is related to memory trace building and temporal window length (I3 versus I6). An increase in CNV amplitude was observed in both conditions, I3 and I6. This increase seems to reflect the building of a memory trace. Moreover, in condition I6, CNV amplitude stopped increasing after the third interval, suggesting that three intervals were clearly a critical limit to the building of this memory trace. This explanation is apparently discrepant with behavioral results, in which there was an enhancement of performance in condition I6 as compared to condition I3. The question is to find out how these additional intervals help subjects improve their performance, as they have already built their memory trace. During the encoding phase, G2 subjects did not exhibit any increase in CNV amplitude after the first interval. Moreover, their behavioral performance was better than that of the other subjects. Apparently, in these G2 subjects, the first time interval was sufficient for creating a memory trace. The other intervals would Figure 4. ERPs over frontal (F5, F3, F1, FZ, F2, F4, F6) and frontalcentral electrodes (FC5, FC3, FC1, FCZ, FC2, FC4, FC6), elicited during tempo encoding in the six-interval and three-interval conditions, averaged separately for G1 subjects (thick lines) and G2 subjects (thin lines).

8 76 M. Pfeuty, R. Ragot, and V. Pouthas then serve merely to check this memory. In other subjects, three intervals seemed to be necessary for performing the same operation, that is, building the memory trace (the next three intervals being used to check it). Thus, the reinforcement of the memory could be due, not only to its building, but also to further checking operations. Interval-based processes relying on attention and working memory appear to be used in the building phase, whereas beat-based processes, certainly more automatic, appear to be used in the checking phase. Conclusion This study suggests that, in tempo perception, two different processes can be differentiated on the basis of the time-course analysis of the CNV. An increase in CNV amplitude reflecting memory trace building would mainly occur during tempo encoding, and a decrease in CNV amplitude reflecting memory trace checking would mainly occur during tempo comparison. REFERENCES Casini, L., & Macar, F. (1996). Can the level of prefrontal activity provide an index of performance in humans? Neuroscience Letters, 219, Drake, C., & Botte, M. C. (1993). Tempo sensitivity in auditory sequences: Evidence for a multiple-look model. Perception & Psychophysics, 54, Gibbon, J., Malapani, C., Dale, C. L., & Gallistel, C. R. (1997). Toward a neurobiology of temporal cognition: Advances and challenges. Current Opinion in Neurobiology, 7, Gratton, G., Coles, M. G. H., & Donchin, E. (1983). A new method for off-line removal of ocular artefacts. Journal of Neuroscience, 18, Harrington, D. L., & Haaland, K. Y. (1999). Neural underpinnings of temporal processing: A review of focal lesion, pharmacological, and functional imaging research. Review Neuroscience, 10, Ivry, R. B. (1996). The representation of temporal information in perception and motor control. Current Opinion in Neurobiology, 6, Ivry, R. B., & Hazeltine, R. E. (1995). Perception and production in temporal intervals across a range of durations: Evidence for a common timing mechanism. Journal of Experimental Psychology: Human Perception and Performance, 21, Keele, S., Nicholetti, R., Ivry, R., & Pokorny, R. (1989). Mechanisms of perceptual timing: Beat-based or interval-based judgments? Psychological Research, 50, Ladanyi, M., & Dubrovsky, B. (1985). CNV and time estimation. International Journal of Neuroscience, 26, Macar, F., Vidal, F., & Casini, L. (1999). The supplementary motor area in motor and sensory timing: Evidence from slow brain potential changes. Experimental Brain Research, 125, McAdam, D. W. (1966). Slow potentials changes recorded from human brain during learning of a temporal interval. Psychonomic Science, 6, Panissal, N. (1998). Le traitement temporel de la duree des intervalles: Mecanisme supramodal et role de la memoire [Time processing of interval durations: Supra-modal mechanisms and the function of memory]. Unpublished doctoral dissertation, Paris University René Descartes. Ro sler, F., Heil, M., & Röder, B. (1997). Slow negative brain potentials as reflections of specific modular resources of cognition. Biological Psychology, 45, Ruchkin, D. S., Berndt, R. S., Johnson, R., Jr., Ritter, W., Grafman, J., & Canoune, H. L. (1997). Modality-specific streams in verbal working memory: Evidence from spatio-temporal patterns of brain activity. Cognitive Brain Research, 6, Ruchkin, D. S., McCalley, M. G., & Glaser, E. M. (1977). Event-related potentials and time estimation. Psychophysiology, 14, Schulze, H. H. (1978). The detectability of local and global displacement in regular rhythmic patterns. Psychological Research, 40, Timsit-Berthier, M. (1984). Variation contingente negative et composantes endogenes du potentiel evoque. Revue d electroencephalographie et de neurophysiologie clinique, 14, Walter, W. G., Cooper, R., Aldridge, V. J., McCallum, W. C., & Winter, A. L. (1964). Contingent negative variation: An electric sign of sensori-motor association and expectancy in the human brain. Nature, 203, (Received March 19, 2001; Accepted July 1, 2002)