REHEARSAL PROCESSES IN WORKING MEMORY AND SYNCHRONIZATION OF BRAIN AREAS

REHEARSAL PROCESSES IN WORKING MEMORY AND SYNCHRONIZATION OF BRAIN AREAS Franziska Kopp* #, Erich Schröger* and Sigrid Lipka # *University of Leipzig, Institute of General Psychology # University of Leipzig, Institute of Linguistics E-mail: fkopp@rz.uni-leipzig.de Abstract In this study we investigated brain activity characteristics of rehearsal processes in working memory using analysis of EEG coherence as a measure of synchronization of brain areas. In a delayed serial recall paradigm we enabled subjects to carry out rehearsal in one condition and disrupted it by irrelevant speech in another condition. Results show that rehearsalspecific changes of coherence duration are located at central electrode sites (C3, Cz, C4) in the Gamma frequency range (35-47 Hz). These changes in synchronization follow our behavioural data that show the classical irrelevant speech effect. Additionally, we found significant coherence changes between frontal and parietal electrodes in the Theta band (4-7.5 Hz) suggesting that rehearsal components can be dissociated from processes reflecting mental effort for retention. Describing psychological processes by the investigation of quantitative external measures has a long tradition and goes back to the beginnings of psychophysics in the 19 th century. During the last few years, however, one can notice an endeavour to find internal characteristics of psychological processes by analyzing brain activity as part of neuropsychological research. This trend results from the great development of imaging techniques on the one hand and on the other hand from the difficulty of describing more complex cognitive processes by traditional behavioural methods such as reaction time or error rate measures. Thus, analyzing brain activity online provides a new and exciting method to investigate temporal and spatial characteristics of so-called higher cognitive processes like memory. In this study we examined a special part of working memory, namely the subvocal rehearsal process. It comprises active maintenance of memory traces and renewal of decaying representations in a short-term store. As one part of the phonological loop in the influential working memory model of Baddeley (1999) rehearsal was proposed on the basis of experiments that examined the disruption of the phonological loop by articulatory suppression or irrelevant speech. In the 1980s and 1990s numerous findings were reported showing which materials disrupt the phonological loop (and hence rehearsal processes) under which conditions. One outcome which is relevant for our experiment showed that presentation of irrelevant speech considerably interrupted phonological rehearsal whereas presentation of unstructured noise did not (Salamé & Baddeley, 1987). Neuropsychological studies using fmri or PET gave evidence for the activation of frontal and parietal brain areas in working memory tasks (e.g. Smith & Jonides, 1997). Other activation studies proved the participation of several brain areas in rehearsal: especially

premotor and prefrontal regions (consistent with the hypothesis that during rehearsal, motor programs are executed like in a recall situation but without overt articulation) (e.g., Fujimaki et al., 1999). However, these structures have to cooperate functionally in order to carry out psychological processes. One special mechanism of this interaction has recently been discussed above all, that is the coupling by synchronization of electrical activity. In our experiment we investigated the synchronization of rehearsal-related brain areas using EEG coherence as an indicator. Coherence gives evidence of the degree of interrelatedness with respect to frequency (this can vary between 0 and 1). As for our experiment, we hypothesized that during rehearsal the duration of high coherence should be long in the prefrontal and premotor regions, and between those two regions. Additionally, disruption of rehearsal is predicted to lead to smaller coherence durations in these areas. For the behavioural data we predicted the classical irrelevant speech effect. Method In our experiment 12 subjects (10 female, 2 male, age 19-29 years, students and professionals) were tested in 3 memory and 3 control conditions. As the paradigm for our memory task we used delayed serial recall (Fig. 1.a). Lists of 5 items were presented sequentially on a PC screen (bisyllabic concrete German nouns similar to the stimuli used by Weiss et al., 2000; they were balanced in frequency and semantic relatedness) with a presentation rate of per item and an ISI of 250 msec. After each list subjects had to retain the items in working memory for an interval of 10 sec. Three question-marks at the end of the retention interval prompted subjects to recall items aloud. Auster Pokal ISI 250 msec. Hügel Retention interval: - quiet - noise - speech 10 sec. Feuer +??? 1.a 1.b Control interval: - quiet - noise - speech 10 sec. Fig. 1. Schematic representation of a trial for the memory tasks (1.a.) and for the nonmemory control tasks (1.b.) +??? The three memory conditions differed during the 10 sec. retention interval: In one condition there was silence in the retention interval intended to enable subjects to subvocally rehearse the items (condition quiet). In another condition irrelevant speech (10 sec. digitalized radio recordings without background noise or music) was presented via headphones during the retention interval to prevent subjects from rehearsing items (condition speech). In the last condition (condition noise) subjects heard white noise during the retention phase in order to expose them to auditory stimulation without disrupting rehearsal (Salamé & Baddeley 1987). Additionally, we tested subjects in 3 control conditions (Fig. 1.b) without memory demands. A fixation cross was presented on an empty screen for 10 sec. During this time, in some trials, a small point appeared for 100 msec. in a region within a few millimetres around the fixation

cross. Upon the presentation of the three question-marks at the end of the 10 sec., subjects had to say if the point had been there or not. This task was used to control attention in the interval. Only "no"-trials were analysed. The 3 control conditions differed like the memory conditions in quiet, noise and speech. During the experiment we recorded EEGs using the 10/20 system with 19 scalp electrodes and the nose as reference point. The Neuroscan system made recordings at a rate of 250 Hz. We used Ag-AgCl electrodes and the modular Easy-cap system. Behavioural data Results and Discussion Fig. 2 shows the percentages of correctly recalled word lists in the memory condition averaged over all subjects. As predicted, noise and quiet did not differ, whereas speech differed significantly both from quiet and from noise. These results show the classical irrelevant speech effect. As for the non-memory control tasks, correct answers reached a level of over 99 % in each condition and conditions did not differ, in line with our predictions. mean % of correctly recalled lists 65 60 55 50 45 40 35 30 quiet noise speech condition Fig. 2. Behavioural results of the memory tasks which show the mean percentages of correctly recalled lists in all three memory conditions. EEG data For the calculation of coherence only artefact-free trials were analyzed with the SpecTrial and SpecPara program based on ARMA models (Schack et al., 1999). This method of analysis allows us to get coherence values with high time and frequency solution. To test our hypotheses, we examined electrode pairs within the frontal region (F7, F3, Fz, F4, F8), within the central region (C3, Cz, C4), between frontal and central electrode sites (combinations of F7, F3, Fz, F4, F8 with C3, Cz, C4) and between frontal and parietal regions (combinations of F3, Fz, F4 with P3, Pz, P4) (Kopp & Sommerfeld, 2000). Based on previous findings reported in the literature we decided to analyze three frequency bands: Beta1 (13-20 Hz), Theta (4-7.5 Hz) and Gamma (35-47 Hz) (Sommerfeld et al., 1999, Tallon-Baudry et al., 1999, Sarnthein et al., 1998). Coherence was computed for a 2 sec. interval within the 10 sec. retention interval (the period between 2 and 4 sec. after the onset of the 10 sec. interval). We computed coherence duration as the sum of all periods of high coherence, i.e., coherence

levels above the threshold of 0.7 within the 2 sec. interval (the threshold resulted from an inspection of coherence histograms). With coherence duration as the dependent variable, statistical tests were carried out separately for each subject because levels of coherence vary considerably between subjects and averaged data do not fulfil requirements of ANOVA. Comparison memory vs. control Single comparisons (t-tests) were carried out for the conditions memory quiet vs. control quiet, memory noise vs. control noise and memory speech vs. control speech to extract the contribution of working memory demand. Fig. 3 shows electrode pairs tested as significant in at least 10 cases of all statistical comparisons. As one can see, several neuronal networks contribute to processing of the memory tasks, showing activity in all three frequency bands. Obviously the Theta band seems to contribute to working memory more than the others and shows differences especially at fronto-parietal electrode pairs. Common to all frequency bands are cooperations in the central area. Our implicit assumption that language-specific effects would occur at left lateral sites was not confirmed. Beta1 (13-20 Hz) Theta (4-7.5 Hz) Gamma (35-47 Hz) Fig. 3. Electrode combinations for the frequency bands Beta1, Theta and Gamma showing differences in coherence duration between memory and control tasks. Solid lines mean significant increases for memory compared to control conditions, the dashed line represents a decrease for memory compared to control. Comparisons within the memory tasks We found that subjects showed very different memory profiles and consequently very different profiles of brain activity: Some of them could recall all 5 items of a list without effort, others had real problems. For some subjects speech had a large disrupting effect on rehearsal, whereas for others speech caused nearly no decrease in performance. Thus it made sense to divide subjects into groups in our analysis of the memory conditions. Two groups were formed based on the strength of the irrelevant speech effect. The first group comprised all subjects with a strong irrelevant speech effect (subjects 3, 5, 8, 9, 10, 11, 12) and the second group consisted of those who were hardly disrupted by irrelevant speech (subjects 1, 2, 4, 6, 7). Statistical analyses of these groups revealed significant differences in the Gamma frequency band at central electrode sites. Fig. 4.a illustrates these results with one representative of each subject group at C3-Cz. As one can see the Gamma coherence duration

decreases significantly in the speech condition for subject 5 who had a strong irrelevant speech effect. Gamma coherence duration does not decrease in subject 7. The results for these two subjects were representative for their respective groups. Additionally, we found significant correlations between recall performance and coherence durations (Gamma band at central electrode sites) only for the subjects with the strong irrelevant speech effect (between 0.26 and 0.43). This rehearsal effect at central electrode positions is consistent with the suggestion that rehearsal is supported by motor programs, as is recall, but without overt articulation. Gamma coherence duration C3-Cz 1200 1100 1000 900 800 700 600 500 400 su b j 7 su b j 5 Theta coherence duration 1200 1100 1000 900 800 700 600 500 400 quiet noise speech 4.a condition 4.b Fz-P3 su b j 4 subj 11 quiet noise speech condition Fig. 4.a. Gamma coherence durations at the central electrode pair C3-Cz for subject 7 (no irrelevant speech effect) and subject 5 (strong irrelevant speech effect) indicating rehearsal. Fig. 4.b. Theta coherence durations at the fronto-parietal electrode pair Fz-P3 in subject 4 (high working memory capacity) and subject 11 (low working memory capacity) indicating mental effort for retention. As mentioned above subjects also varied concerning their working memory capacity. Based on a measure of working memory capacity, digit span, we divided subjects again into 2 groups: one with subjects of high capacity (subjects 4, 5, 8, 10, 12) and the other with subjects of low capacity (subjects 3, 9, 1, 2, 6, 7, 11). Statistical comparisons of these groups revealed significant differences in fronto-parietal electrode combinations in the Theta frequency band as shown in Fig. 4.b (for two typical members of each group) for the electrode pair Fz-P3. Subjects with low capacity, such as subject 11, have significantly longer coherence durations overall compared to subjects with high capacity (see subject 4). Regarding differences between memory conditions, results show that high-capacity subjects show an increase of Theta coherence durations in noise and in speech compared to quiet, that means an increase with task difficulty. Subjects with low capacity do not show any differences between memory conditions. Fronto-parietal coupling of brain activation especially in the Theta frequency range has been described before as an indicator for mental effort for retention (Sommerfeld et al., 1999). This could be reflected in our data as well. Subjects with high working memory capacity, i.e., with more resources, use less effort to retain 5 items in working memory and they increase their effort with increasing memory demands. However, for low capacity

subjects task requirements are high from the outset and remain high for all memory conditions. In sum, the EEG coherence analysis proved to be a very useful tool to show which brain areas cooperate during rehearsal processes in a working memory task. Furthermore, group differences in recall performance corresponded to differences in the EEG coherence measures. Our study also demonstrated that coherence analysis is a useful method to dissociate different components of working memory, i.e., rehearsal processes and mental effort for retaining items. Acknowledgement We would like to thank Esther Herrmann and Frank-Michael Schleif for their assistance in data collection and analysis and Dr. Erdmute Sommerfeld for her advice in all matters of coherence. This work was supported by Deutsche Forschungsgemeinschaft. References Baddeley, A.D. (1999). Working memory: The multiple-component model. In: Miyake, A. & Shah, P (Eds.) Models of working memory - mechanisms of active maintenance and executive control. Cambridge University Press. Fujimaki, N., Nielsen, M., Hayakama, T., Kato, M. & Mijauchi, S. (1999). Neural activity dependent on phonological demands in a verbal working memory task. 5 th International Conference on Functional Mapping of the Human Brain, Düsseldorf. Kopp, F. & Sommerfeld, E. (2000). Dissociation of control processes in working memory from visual and motor processes by EEG coherence. Brain Topography, 12, 295. Salamé, P. & Baddeley, A.D. (1987). Noise, unattended speech and short-term memory. Ergonomics, 30, 1185-1193. Sarnthein, J., Petsche, H., Rappelsberger, P., Shaw, G.L. & von Stein, A. (1998). Synchronization between prefrontal and posterior association cortex during human working memory. Proc. Natl. Acad. Sci. USA, 95, 7092-7096. Schack, B., Grieszbach, G. & Krause, W. (1999). The sensitivity of instantaneous coherence for considering elementary comparison processing. Part I: The relationship between mental activities and instantaneous EEG coherence. International Journal of Psychophysiology, 31, 219-240. Smith, E.E. & Jonides, J. (1997). Working memory: a view from neuroimaging. Cognitive Psychology, 33, 5-42. Sommerfeld, E., Hensel, A. & Hildebrandt (Simmel), A. (1999). Cooperation of frontal and parietal brain areas as a function of cognitive training. In: Killeen, P.R. & Uttal, W.R. (Eds.) Fechner Day 99: The end of 20 th century psychophysics. Proceedings of the fifteenth annual meeting of the International Society for Psychophysics. Tempe, AZ, USA: The International Society for Psychophysics, 356-361. Tallon-Baudry, C., Kreiter, A. & Bertrand, O. (1999). Sustained and transient oscillatory responses in the gamma and beta bands in a visual short-term memory task in humans. Visual Neuroscience, 16, 449-459. Weiss, S., Müller, H.M. & Rappelsberger, P. (2000). Theta synchronization predicts efficient memory encoding of concrete and abstract nouns. NeuroReport, 11 (11), 2357-2361.