Cognitive Brain Research 6 1997 83–94
`.
`Ž
`
`Research report
`Event-related desynchronization in the alpha band and the processing of
`semantic information
`W. Klimesch ), M. Doppelmayr, T. Pachinger, H. Russegger
`Department of Physiological Psychology, UniÕersity of Salzburg, Hellbrunnerstr. 34, A-5020 Salzburg, Austria
`Accepted 30 April 1997
`
`Abstract
`
`The hypothesis was tested whether event-related power shifts in the upper alpha band are specifically related to semantic memory
`processes. In Expt. 1 subjects had to judge whether pairs of sequentially presented words W1-W2 were semantically congruent. In the
`Ž
`.
`following experiments subjects were presented the W1 words of Expt. 1 and were asked to perform a free association task in Expt. 2 and
`a cued recall task in Expt. 3. It is assumed that semantic memory demands dominate in Expt. 1, whereas working memory demands
`dominate in Expt. 3 and that Expt. 2 takes an intermediate position with respect to both types of task demands. A significant task-related
`power change that responds selectively to semantic processing demands was found for the upper alpha band and over the left side of the
`scalp. The lower alpha band, on the other hand, most likely reflects unspecific processing demands such as attention. A more general
`interpretation of these findings is that different cognitive processes such as semantic memory, perceptual encoding and attentional
`processes are reflected by band power changes in different and rather narrow frequency bands over localized regions in the brain. q 1997
`Elsevier Science B.V.
`
`Keywords: Lower alpha; Upper alpha; Desynchronization; Semantic memory; Theta; Episodic memory
`
`1. Introduction
`
`The finding that alpha desynchronizes becomes sup-
`Ž
`pressed in response to a variety of different
`tasks is
`.
`known since the early days of EEG research e.g. 2 . It
`Ž
`w x.
`was and still is generally believed that visual or other
`Ž
`.
`Ž
`sensory
`task demands,
`including visual attention
`cf.
`.
`Ž
`24,25 , are the primary factors that lead to a suppression
`x.
`w
`of the alpha rhythm. More recent research, however, has
`revealed a much more complex picture.
`A better understanding of the functional meaning of the
`alpha rhythm was provided by a new method, which
`Pfurtscheller and his coworkers termed ERD event-related
`Ž
`desynchronization 28 . This method allows to calculate
`x.
`w
`the percentage of event-related power changes for different
`frequency bands. One important finding obtained with this
`method is that the degree and topography of desynchro-
`nization show large and reliable differences between the
`lower and upper alpha band. As an example, desynchro-
`
`) Corresponding author. Fax: q43 662 8044 5126; E-mail:
`Wolfgang.Klimesch@sbg.ac.at
`
`0926-6410r97r$17.00 q 1997 Elsevier Science B.V. All rights reserved.
`
`nization in the lower alpha band is topographically
`widespread without any clear localization, whereas in the
`upper alpha band, desynchronization tends to be more
`localized at those areas which play an important role in
`processing of a particular type of task. For a variety of
`motor tasks Pfurtscheller and colleagues e.g. 23,29,30
`Ž
`x.
`w
`have demonstrated that desynchronization in the upper
`alpha band of about 10–12 Hz
`is localized over the
`Ž
`.
`respective area of the motor cortex over the left or right
`side of the scalp. The fact that the two alpha bands show
`strikingly different results supports the proposal that there
`is no single alpha rhythm but instead a variety of different
`alpha frequencies 20,21 .
`w
`x
`The method of ERD and the distinction between differ-
`ent alpha bands also proved useful for the analysis of
`cognitive processes. As an example, a warning signal
`preceding the presentation of an imperative stimulus causes
`a strong short-lasting synchronization followed by a desyn-
`chronization which most interestingly appears in the lower
`alpha band only 13–15 . In Expt. 1 of Klimesch et al.
`w
`x
`x13 , subjects were required to read a series of words. The
`w
`only manipulation was the variation of the time interval
`between the warning signal and the presentation of a word.
`
`Wave Neuroscience EX2004
`PeakLogic v. Wave Neuroscience
`IPR2022-01550
`
`

`

`84
`
`)
`(
`W. Klimesch et al.rCognitiÕe Brain Research 6 1997 83–94
`
`For the upper alpha band only the presentation of a word –
`but not the warning signal – leads to a localized desyn-
`chronization at posterior sites 11,13–15,17,18 . In con-
`w
`x
`trast, the lower alpha band responds to the presentation of
`both,
`the warning signal and the imperative stimulus.
`These and similar results cf. Expt. 2 in Klimesch et al.
`Ž
`have led to suggest the hypothesis that the
`13,15,17
`x.
`w
`lower alpha band is related to general task demands such
`as attention see also Crawford et al. 5 , whereas the
`Ž
`w x.
`upper alpha band is primarily associated with specific task
`demands such as the visual andror semantic processing of
`the imperative stimulus.
`This latter aspect of the suggested hypothesis was tested
`in a more recent experiment which was designed to com-
`pare the effects of episodic memory and semantic process-
`ing demands 16 . In this experiment we used a design that
`w
`x
`already proved useful to distinguish semantic from episodic
`memory processes
`see Expt. 4 in Kroll and Klimesch
`Ž
`x.19 . The experimental design consisted of two parts.
`w
`Subjects first performed a semantic congruency task in
`which they had to judge whether or not the sequentially
`presented words of concept-feature pairs such as ‘‘eagle-
`Ž
`claws’’ or ‘‘pea-huge’’ are semantically congruent. Then,
`.
`without prior warning,
`they were asked to perform an
`episodic recognition task. This was done in an attempt to
`prevent subjects from using semantic encoding strategies
`and thus to increase episodic memory demands. In the
`episodic task, the same concept-feature pairs were pre-
`sented together with new distractors generated by repair-
`Ž
`ing known concept-feature pairs . Now subjects had to
`.
`judge whether or not a particular concept-feature pair was
`already presented during the semantic task. From the re-
`sults found in Kroll and Klimesch 19 we know that the
`w
`x
`episodic task showing longer RTs and a larger percentage
`Ž
`of incorrect responses
`is much more difficult than the
`.
`semantic task. This latter feature of the design is important
`for our central prediction which is that upper alpha desyn-
`chronizes selectively in response to semantic task de-
`mands. When considering the fact that alpha desynchro-
`nizes with increasing task difficulty, we would expect the
`opposite effect, which is a stronger desynchronization
`during the episodic task. Thus, if task difficulty would be
`the only factor which is reflected by an event-related
`decrease in the upper alpha band reflected by an increase
`Ž
`in ERD we would expect the most pronounced increase in
`.
`ERD during the presentation of the feature in the episodic
`task. This, however, is not the case. The results indicate
`that in spite of the fact that the semantic task is easier than
`the episodic task, the upper alpha band shows a signifi-
`cantly stronger desynchronization during the processing of
`the semantic task.
`The present experiment is designed to further test the
`hypothesis whether the upper alpha band is specifically
`related to the processing of semantic information. The
`experimental design consists of three different tasks termed
`Ž
`Expt. 1, 2 and 3 in the following in which semantic
`.
`
`processing demands are varied. The same sample of sub-
`jects was used in all of the three experiments. In a similar
`way as in the semantic task of Klimesch et al. 16 , in
`w
`x
`Expt. 1 subjects had to judge whether pairs of sequentially
`presented words W1-W2 were semantically congruent. In
`Ž
`.
`the following experiments subjects were presented the W1
`words of Expt. 1 and were asked to perform a free
`association task in Expt. 2 and a cued recall task in Expt.
`3.
`
`Semantic memory is pure long-term memory 7 . The
`w x
`knowledge about semantic relationships like ‘‘an eagle has
`claws’’ is known since childhood and represents ‘‘pre-ex-
`perimental knowledge’’. On the other hand, ‘‘experimental
`knowledge’’ is tested in working memory tasks where
`subjects have to retrieve information that was presented
`earlier in the experiment 1 . Thus, semantic memory
`w x
`demands vary from high in the semantic congruency task
`of Expt. 1 to low in the cued recall
`task of Expt. 3.
`Conversely, working memory demands are high in Expt. 3
`and low in Expt. 1. Expt. 2 takes an intermediate position.
`In a similar way as in Expt. 1 semantic relationships may
`be used to give a response. However, in contrast to Expt.
`1, subjects may use their working memory either to gener-
`ate new words or to retrieve the respective W2 words. The
`term ‘‘free association task’’ is used for Expt. 2 because
`subjects were instructed explicitly to report any word that
`comes into their mind.
`Given this description of task demands, the following
`predictions can be tested:
`Ž .i The semantic congruency task of Expt. 1 can be
`carried out only after the presentation of the second word
`of a pair. Thus, for the upper alpha band we expect the
`strongest task-related desynchronization in the poststimu-
`lus period following the presentation of the W2 word.
`Ž .ii
`In contrast to Expt. 1, semantic memory demands
`are low in Expt. 2 and 3. Thus, as compared to the W2
`words of Expt. 1, desynchronization in the upper alpha
`band will be smaller in Expt. 2 and 3.
`.iii
`In all of the three experiments, general processing
`Ž
`demands, such as attention and effort, will increase from
`the beginning to the end of a task. For the lower alpha
`band we,
`thus, expect a gradual, stepwise increase in
`desynchronization from the beginning of a trial until a
`response is given.
`
`2. Materials and methods
`
`2.1. Subjects
`
`Subjects were 12 right-handed students 7 males, 5
`Ž
`females who participated voluntarily in the experiment.
`.
`Their mean age was 23.7 years with a range of 20–31
`years. Handedness was controlled by asking the subjects
`about the hand they use in different tasks such as handwrit-
`ing, throwing a ball, etc. A prospective subject was consid-
`
`

`

`)
`(
`W. Klimesch et al.rCognitiÕe Brain Research 6 1997 83–94
`
`85
`
`ered right-handed if hershe indicated to use the right hand
`for all of these different tasks.
`
`2.2. Materials
`
`A set of 192 feature and concept words was used as
`stimuli. Half of the words belonged to the category ‘‘liv-
`ing’’ birds, fruits, vegetables , the other half to ‘‘non-liv-
`Ž
`.
`ing’’ vehicles, clothes, weapons . For a detailed descrip-
`Ž
`.
`tion of a similar set of stimuli, including word norms, see
`Kroll and Klimesch 19 .
`w
`x
`The words appeared at the center of a computer monitor
`and were 0.7 cm in height. The length of a word with 10
`letters was 5.7 cm. Subjects sat at a distance of 90 cm
`from the monitor.
`
`2.3. Design
`
`In Expt. 1 subjects had to perform a semantic congru-
`ency task on sequentially presented pairs of words W1,
`W2 by responding ‘‘yes’’ to a congruent and ‘‘no’’ to an
`incongruent pair e.g., W1 sclaws, W2 seagle; W1 s
`Ž
`blue, W2 scanary . Subjects were instructed to report as
`.
`quickly as possible but not before a response signal cf.
`Ž
`Fig. 1 was presented. The W1 words are also called
`.
`‘‘feature words’’, whereas the W2 words represent ‘‘con-
`cept words’’. Feature words were paired with concept
`words W2 words
`in a way that half of the pairs were
`Ž
`.
`congruent, the other half incongruent. Subjects were in-
`formed that word pairs representing a ‘‘feature’’ and a
`‘‘concept’’ e.g., ‘‘claws-eagle’’ will be presented. They
`Ž
`.
`were asked to give a positive response if the pair is
`semantically congruent and a negative response if the pair
`is incongruent e.g., ‘‘blue-canary’’ .
`Ž
`.
`In Expt. 2 and 3 only half of the W1 words were
`presented. These were those feature words which in Expt.
`1 were paired with congruent concept words. The experi-
`mental design is shown in Fig. 1.
`In Expt. 2 subjects were asked to carry out a free
`association task. They were instructed to report any word
`that would come into their minds as quickly as possible
`but after a response signal was presented. Subjects knew
`that feature words of Expt. 1 – arranged in a different
`sequence – will be presented. They were asked to avoid
`repetitions, i.e., responding with the same association more
`than once.
`Finally, in Expt. 3 and without prior warning a cued
`recall task was performed. Subjects were informed that
`half of the feature words of Expt. 1 – arranged in a
`different sequence – will be presented again. But now they
`were asked to remember and report that concept word that
`was paired with the respective feature word in the seman-
`tic task of Expt. 1. Again, subjects were asked to wait with
`their response until a response signal appeared on the
`screen.
`
`Fig. 1. In Expt. 1 subjects had to judge whether pairs of sequentially
`presented feature words W1 and concept words W2 were semantically
`Ž
`.
`Ž
`.
`congruent. In Expt. 2 subjects were asked to respond with a free
`association to W1, and in Expt. 3, W1 served as a cue to recall the
`respective concept word W2 . A warning signal WS preceded the
`Ž
`.
`Ž
`.
`presentation of W1. Test intervals t1–t5 represent time periods of 500 ms
`each.
`
`The main reason for using only those feature words
`which were congruent in Expt. 1 was to avoid that the type
`of response in Expt. 1 which represents episodic informa-
`Ž
`tion interferes or interacts with the type of response in
`.
`Expt. 2 and 3 in an uncontrolled way. As an example,
`when trying to remember the corresponding concept word
`in Expt. 3 subjects could try to remember whether a
`presented feature word elicited a yes or no response in
`Expt. 1. A strategy like this would prevent subjects from
`retrieving the semantic relationship between the feature
`and concept word. The use of congruent feature words in
`Expt. 2 and 3 makes it, thus, more likely that a subject’s
`response is based on task relevant semantic instead of
`irrelevant episodic information.
`As indicated by Fig. 1, the EEG is analyzed in time
`intervals of 500 ms each, preceding and following the
`presentation of a word. These time intervals are termed
`t1–t5. Interval t1 represents the prestimulus period. During
`t2 and t4 in Expt. 1 the presented W1 or W2 word is
`Ž
`.
`encoded, whereas in t3 and t5 in Expt. 1 the primary
`Ž
`.
`focus is on the cognitive processing of the respective word.
`Semantic congruency judgments are easy and,
`thus,
`quite fast to perform. As the analysis of reaction times
`.RTs has shown, the average semantic verification time –
`Ž
`as measured from the onset of the W2 word – is less than
`1 s cf. Expt. 2a in Kroll and Klimesch 19 . In addition,
`Ž
`x.
`w
`the results of Expt. 2b in Kroll and Klimesch 19 have
`w
`x
`shown that under otherwise comparable conditions lexi-
`.
`Ž
`cal verification time the time needed to distinguish a word
`Ž
`from a non-word is 677 ms. Earlier performed experi-
`.
`ments 10 have indicated that about 200 ms are needed to
`w
`x
`carry out the type of motor response that was required in
`these tasks i.e., moving the right finger and pressing the
`Ž
`respective response key . This means that
`in the first
`.
`poststimulus interval of 0–500 ms t2 or t4 respectively a
`Ž
`.
`word is lexicographically processed, whereas semantic ver-
`ification or processing takes place primarily during the
`second poststimulus interval
`t3 or t5 respectively .
`Ž
`.
`
`

`

`86
`
`2.4. Procedure
`
`)
`(
`W. Klimesch et al.rCognitiÕe Brain Research 6 1997 83–94
`
`All of the words were presented for 250 ms on a
`computer monitor. In all of the experiments, the length of
`a single trial epoch was 8 s. A brief warning signal 3000
`Ž
`.
`Ž
`Hz, lasting for 250 ms appeared 1000 ms before a W1
`.
`word was presented. The structure of single trials is shown
`in Fig. 1.
`In an attempt to avoid artifacts, subjects were asked to
`maintain fixation by looking at the middle of the screen as
`soon as an acoustic warning signal appeared. As Fig. 1
`illustrates, the response signal question mark appeared
`Ž
`.
`1500 ms after presentation onset of a W2 word in Expt. 1
`Ž
`.
`or a W1 word in Expt. 2 and 3 . This procedure was also
`Ž
`.
`used to avoid response artifacts and to obtain poststimulus
`processing times which are of comparable duration in all
`of the three conditions. This was done even at the risk of a
`slight increase in working memory demands at or after t5
`or t3 respectively.
`Subjects responded verbally by saying ‘‘yes’’ or ‘‘no’’
`in Expt. 1, by reporting an association in Expt. 2 and by
`reporting a concept word in Expt. 3. All of the responses
`were recorded by the experimenter. Each experiment was
`preceded by a series of four training trials.
`
`2.4.1. Apparatus
`EEG signals were amplified by a 32-channel biosignal
`amplifier system frequency response: 0.16–30 Hz , sub-
`Ž
`.
`jected to an anti-aliasing filter bank cut-off frequency: 30
`Ž
`Hz, 110 dBroctave and were then converted to a digital
`.
`format via a 32-channel ArD converter. Sampling rate
`was 128 Hz. The data were processed by the B.E.S.T.
`system of Grossegger and Drbal which is based on a 486
`PC. During data acquisition, EEG signals were displayed
`on-line on a high resolution monitor and stored on disk.
`
`2.4.2. Recordings
`A set of 22 silver electrodes, attached with a glue paste
`to the scalp, was used to record EEG signals. Ten elec-
`trodes were placed according to the international electrode
`10–20 placement system, at F3, F4, C3, C4, T3, T4, P3,
`Ž
`.
`P4, O1 and O2. From the remaining 12 electrodes, four
`electrodes were placed over parieto-occipital areas PO3,
`Ž
`PO1, PO2, PO4 , four electrodes were placed over centro-
`.
`parietal areas CP5, CP1, CP2, CP6 , and four electrodes
`Ž
`.
`were placed over fronto-central areas FC5, FC1, FC2,
`Ž
`FC6 . For the placement of these additional electrodes see,
`.
`e.g., Herrmann et al. 6 .
`w x
`In addition to the 22 electrodes described above, two
`mastoid electrodes termed A1 and A2 were attached to
`Ž
`.
`the left and right ear. Furthermore, the electrooculogram
`EOG was recorded from two pairs of leads in order to
`Ž
`.
`register horizontal and vertical eye movements.
`All data were recorded monopolarly against a common
`reference placed on the nose. In order to eliminate the
`effects of the nose reference as well as other types of
`
`artifacts, the EEG recordings were corrected by subtracting
`the arithmetically averaged mastoid recordings A1 q
`Ž
`.A2 r2 from all of the monopolar recordings.
`All of the epochs in each of the three experiments were
`checked individually for artifacts eye blinks, horizontal
`Ž
`and vertical eye movements, muscle artifacts, etc. by
`.
`visual inspection. Only epochs with a correct yes or no
`response were used for data analysis. The percentage of
`epochs that were excluded from data analysis was 5.9%
`Ži.e., no response to a congruent pair, or yes response to an
`incongruent pair in Expt. 1, 12.9% missing association or
`.
`Ž
`associations reported more than once
`in Expt. 2 and
`.
`12.1% missing response or concept word reported more
`Ž
`than once in Expt. 3.
`.
`
`2.4.3. The calculation of eÕent-related changes in band
`power
`The data for each epoch and each of the 22 channels
`were digitally band-pass filtered, squared in order to
`Ž
`obtain simple power estimates and averaged separately for
`.
`each experimental condition and for each subject. Based
`on these data, event-related changes in band power were
`calculated in using a procedure which was originally pro-
`posed by Pfurtscheller and Aranibar 28 who coined the
`w
`x
`term event-related desynchronization or ERD. ERD is
`defined as the percentage of decrease or increase in band
`power during a test interval as compared to a reference
`interval: ERD % s band power, reference interval y
`wŽ
`.
`band power, test interval r band power, reference inter-
`Ž
`.x
`Ž
`val =100. For a detailed description see e.g. 31 . Note
`.4
`w
`x
`that positive ERD values indicate a state of desynchroniza-
`tion or power suppression. Negative ERD values, on the
`other hand, reflect a state of synchronization or increase in
`band power.
`In the present study, an interval of 1000 ms beginning
`Ž
`2500 ms before and ending 1500 ms before the presenta-
`tion of the feature word was used as reference interval cf.
`.
`Ž
`Fig. 1 . The test intervals consist of time intervals of 500
`.
`ms preceding and following the presentation of a word cf.
`Ž
`the time intervals t1–t5 in Fig. 1 .
`.
`After rejecting artifacts, an average of 147, 56 and 63
`epochs remained for Expt. 1, 2 and 3 respectively.
`
`2.4.4. The indiÕidual determination of frequency bands
`The frequency windows for the theta and alpha bands
`were determined individually for each subject i by using
`mean peak frequency f i of the dominant EEG frequency
`Ž .
`in the alpha band for all recording sites as an anchor point.
`Mean peak frequency f i was calculated over the entire
`Ž .
`epoch of 8 s. In using f i as individual anchor point, four
`Ž .
`different frequency bands with a bandwidth of 2 Hz each
`were defined:
`f i -6 to f i -4 ;
`f i -4 to f i -2 ;
`f i -2
`Ž Ž .
`.
`Ž Ž .
`. Ž Ž .
`.
`Ž Ž .
`. Ž Ž .
`.
`to f i q2 . The ERD was calculated within
`to f i ; f i
`Ž Ž .
`Ž .
`Ž .
`.
`these individually determined frequency bands which are
`termed theta,
`lower-1 alpha,
`lower-2 alpha, and upper
`alpha. The averaged mean frequency f i was 10.0 Hz.
`Ž .
`
`

`

`)
`(
`W. Klimesch et al.rCognitiÕe Brain Research 6 1997 83–94
`
`87
`
`Thus, averaged over the entire sample of 12 subjects, the
`following cut-off points were obtained: 4–6 Hz, 6–8 Hz,
`8–10 Hz, 10–12 Hz.
`With frequency bands of 2 Hz, epochs down to a length
`of 0.5 s can still be analyzed. However, the effects of
`spectral leakage i.e.,
`the influence of neighboring fre-
`Ž
`quency bands becomes the more severe, the shorter the
`.
`epoch length is. One way to compensate for this effect is
`to use a large number of epochs over which power esti-
`mates are averaged. By averaging, the distorting effects of
`neighboring frequency bands decrease and the accuracy of
`power estimates increases. This is for the same reason why
`the accuracy of any estimate increases with increasing
`sample size. Because only the averaged data of at least 56
`Ž
`trials are used for statistical analyses, the reported results
`.
`are based on power estimates that can be considered
`sufficiently reliable.
`With respect to the naming of the frequency bands it
`should be emphasized that the frequency band of 6–8 Hz,
`
`which is termed ‘‘lower-1 alpha’’, shows only a marginal
`overlap with the traditional alpha band of 7–13 Hz. How-
`ever, as the results reported below will show, this fre-
`quency band exhibits effects that are quite similar to the
`lower-2 alpha which ranges from 8 to 10 Hz. As the
`lower-2 alpha band and in contrast to the theta band, the
`lower-1 alpha band desynchronizes with increasing task
`demands. Thus, for simplicity, we use the term lower-1
`alpha even for this slow frequency range.
`This method of adjusting frequency bands individually
`before analyzing ERD is important because of the follow-
`ing two findings:
`i alpha frequency varies as a function
`Ž .
`of several factors such as age, memory performance and
`attentional demands cf. 11,14 and the summary in 8,9 ;
`Ž
`x.
`w
`x
`w
`the frequency of the transition region between a task-
`Ž .ii
`related decrease in band power as an indicator for the
`Ž
`alpha band and a task-related increase in band power as
`.
`Ž
`an indicator for the theta band shows also a tendency to
`.
`vary as a function of alpha frequency see e.g. 18 . It is,
`Ž
`x.
`w
`
`Table 1
`
`Time
`ANOVA results: Expt. 1, semantic task
`df numerator
`4
`Ž
`.
`df denom.
`44
`Ž
`.
`Theta
`3.13
`0.55
`24.32
`0.43
`39.92
`0.58
`5.28
`0.43
`
`Lower-1 alpha
`
`Lower-2 alpha
`
`Upper alpha
`
`F
`
`e
`
`F
`
`e
`
`F
`
`e
`
`F
`
`e
`
`Location
`
`4
`44
`2.00
`0.41
`1.47
`0.57
`4.00
`0.57
`11.67
`0.59
`
`a
`
`b
`
`4
`44
`1.28
`0.46
`1.08
`0.57
`4.97 a
`0.55
`7.31
`0.50
`
`b
`
`4
`44
`2.73
`0.44
`1.16
`0.55
`11.05
`0.60
`23.83
`0.61
`
`b
`
`b
`
`Hemi
`
`1
`11
`0.54
`–
`1.44
`–
`1.51
`–
`2.31
`–
`
`1
`11
`2.18
`–
`0.01
`–
`0.07
`–
`1.51
`–
`
`1
`11
`0.00
`–
`0.91
`–
`1.25
`–
`0.34
`–
`
`T =L
`
`16
`176
`3.62
`0.29
`2.08
`0.13
`3.92
`0.17
`4.60
`0.15
`
`8
`88
`4.38
`0.47
`2.50
`0.42
`2.67
`0.39
`3.04
`0.36
`
`8
`88
`8.35
`0.39
`8.52
`0.34
`3.34
`0.35
`6.19
`0.30
`
`b
`
`a
`
`a
`
`b
`
`a
`
`b
`
`b
`
`a
`
`b
`
`T =H
`
`L =H
`
`T =L =H
`
`4
`44
`0.17
`0.68
`0.10
`0.43
`0.66
`0.40
`7.63
`0.57
`
`2
`22
`3.96
`0.96
`0.64
`0.82
`0.22
`0.60
`2.40
`0.95
`
`2
`22
`0.02
`0.74
`0.55
`0.79
`4.75
`0.83
`2.86
`0.82
`
`b
`
`a
`
`a
`
`4
`44
`0.90
`0.49
`0.26
`0.63
`0.45
`0.44
`3.20
`0.65
`
`4
`44
`0.07
`0.62
`0.89
`0.55
`1.98
`0.60
`3.00
`0.54
`
`4
`44
`1.40
`0.75
`1.71
`0.56
`2.59
`0.62
`4.64
`0.59
`
`a
`
`a
`
`a
`
`b
`
`a
`
`16
`176
`0.95
`0.34
`2.08
`0.32
`2.69
`0.23
`5.15
`0.30
`
`8
`88
`0.72
`0.42
`2.58
`0.40
`2.59
`0.35
`3.40
`0.31
`
`8
`88
`1.59
`0.38
`0.96
`0.51
`1.63
`0.47
`2.51
`0.42
`
`b
`
`b
`
`a
`
`b
`
`a
`
`b
`
`ANOVA results: Expt. 2, free association
`df numerator
`2
`Ž
`.
`df denom.
`22
`Ž
`.
`Theta
`1.41
`0.81
`16.85
`0.65
`12.20 b
`0.78
`4.65
`0.77
`
`Lower-1 alpha
`
`Lower-2 alpha
`
`Upper alpha
`
`F
`
`e
`
`F
`
`e
`
`F
`
`e
`
`F
`
`e
`
`ANOVA results: Expt. 3, cued recall
`df numerator
`2
`Ž
`.
`df denom.
`22
`Ž
`.
`Theta
`1.94
`0.84
`2.73
`0.95
`36.34
`0.89
`2.84
`0.88
`
`Lower-1 alpha
`
`Lower-2 alpha
`
`Upper alpha
`
`F
`
`e
`
`F
`
`e
`
`F
`
`e
`
`F
`
`e
`
`T, time; L, location; H, Hemi recording sites over the left and right sides of the scalp .
`.
`Ž
`a 5%; b 1% level of significance.
`
`

`

`88
`
`)
`(
`W. Klimesch et al.rCognitiÕe Brain Research 6 1997 83–94
`
`as a
`important
`thus,
`to use alpha peak frequency f i
`Ž .
`common reference point for adjusting different frequency
`bands not only for the alpha, but theta range as well.
`
`2.4.5. Statistical analyses
`Three different
`types of ANOVAs were calculated.
`First, 3-factorial ANOVAs were calculated to compare the
`W1 words of congruent pairs with the W1 words of
`incongruent pairs of Expt. 1. Second, 3-factorial ANOVAs
`were calculated for comparisons within each of the three
`experiments. Third, 4-factorial ANOVAs were computed
`to analyze effects that are due to differences between the
`experiments.
`In order to check whether the W1 words in congruent
`and incongruent pairs represent homogeneous subsamples,
`ERD data were subjected to 3-factorial ANOVAs. This
`was done separately for each of the four frequency bands
`and for each of the first three time intervals t1, t2, t3 . The
`Ž
`.
`factors and their levels are: Kong: W1 words of congruent
`pairs versus W1 words of incongruent pairs; Location:
`frontal F3, F4, FC5, FC6 , central C3, C4, FC1, FC2,
`Ž
`.
`Ž
`CP1, CP2 , parietal P3, P4, CP5, CP6 , temporal T3, T4
`.
`Ž
`.
`Ž
`.
`and occipital O1, O2, PO3, PO1, PO2, PO4 leads; and
`Ž
`.
`Hemi: recording sites over the left and right
`l, r side of
`Ž
`.
`the scalp.
`For each of the three experiments and each of the four
`frequency bands, ERD data were subjected to 3-factorial
`ANOVAs. The factors are Time, Location, and Hemi. For
`factor Time the levels are intervals of 500 ms preceding
`and following the presentation of a word t1, t2, t3, t4 and
`Ž
`t5 in Expt. 1; t1, t2 and t3 in Expt. 2 and 3 . The levels of
`.
`Location and Hemi were already described above. The
`results of the respective ANOVAs are summarized in
`Table 1.
`For the analysis of differences between the experiments
`4-factorial ANOVAs were calculated for each frequency
`band. The factors Location, Hemi and Time with only
`Ž
`three levels: t1, t2, and t3 were identical with the respec-
`.
`tive factors of the 3-factorial ANOVAs. The only addi-
`tional factor is Task with three levels: semantic, free
`Ž
`association, and cued recall task .
`.
`The Greenhouse–Geisser procedure was used to com-
`pensate for violations of sphericity or circularity. For
`repeated measurement factors with more than two levels,
`the epsilon factor e and the adjusted tail probabilities
`Ž
`.
`will be reported.
`
`3. Results
`
`3.1. BehaÕioral data
`
`In Expt. 1, an average of 5.9% of the responses were
`incorrect yes response to an incongruent or no response to
`Ž
`a congruent pair . In Expt. 2, in 22.5% of the cases a
`.
`subject associated the correct concept word, in 31.8% of
`
`the cases any concept or feature word that was presented
`in Expt. 1. An average of 32.8% were new associations.
`No association or repeated associations were observed in
`12.8% of the cases. In Expt. 3, the percentage of correctly
`remembered concept words was 29.3%
`
`3.2. Experiment 1
`
`None of the 12 3-factorial ANOVAs which were calcu-
`lated to check for differences between the group of W1
`words of congruent and that of incongruent pairs showed
`significant results at or beyond the 5% level
`in which
`Ž
`.
`factor Kong either as main effect or interaction is in-
`Ž
`.
`volved. This finding demonstrates that the W1 words of
`congruent and incongruent pairs can be considered a ho-
`mogeneous sample. Therefore, the data for congruent and
`incongruent pairs were pooled in all of the analyses re-
`ported below. This finding is important because only those
`W1 words were used in Expt. 2 and 3 which belonged to
`congruent pairs in Expt. 1 see Section 2 .
`Ž
`.
`For the theta band, only one significant result was
`found as the summary of results in Table 1 shows. Inspec-
`tion of the respective means of the significant interaction
`Time =Location reveals that only at occipital recording
`sites and only during t2 stimulus onset of W1 and t4
`Ž
`.
`stimulus onset of W2 , a pronounced synchronization
`Ž
`.
`increase in power was found.
`Ž
`.
`For the lower-1 alpha band the only significant finding
`is a main effect for Time. The respective means are plotted
`in Fig. 2 and show a gradual increase in desynchronization
`´
`from t1 to t5. According to the Scheffe test for pairwise
`comparisons the critical difference at the 5% level is 10.2.
`As compared to t1, the increase in ERD is significant at
`and beyond t3.
`The significant main effect for Time in the lower-2
`alpha band shows a similar result cf. Fig. 2 . In addition,
`Ž
`.
`factor Location indicating that the amount of desynchro-
`Ž
`nization is more pronounced at posterior as compared to
`anterior
`recording sites
`and the interactions Time =
`.
`Location and Time =Location =Hemi reached signifi-
`cance. The 3-way interaction indicates a left hemispheric
`advantage stronger desynchronization over the left side of
`Ž
`the scalp which can be observed during t2–t4 but not in
`.
`Ž
`t1 and which is particularly large at anterior
`including
`Ž
`.
`temporal
`recording sites. For the levels of the significant
`.
`´
`main effect of Time, the Scheffe test for pairwise compar-
`isons yields a critical difference of 10.6 at the 5% level. As
`compared to t1, the increase in ERD is significant at and
`beyond t3.
`In the upper alpha band all but one variance sources
`reached significance. The significant main effect for Time
`´
`is depicted in Fig. 2a. The Scheffe test for pairwise
`comparisons yields a critical difference of 18.9 at the 5%
`level. As indicated by the respective arrow in Fig. 2a, the
`increase in poststimulus ERD is significant at t5 only.
`Because the main effects as well as the three significant
`
`

`

`)
`(
`W. Klimesch et al.rCognitiÕe Brain Research 6 1997 83–94
`
`89
`
`3.3. Experiment 2
`
`The theta band shows two significant interactions, Time
`=Location and Time =Hemi. Inspection of the respective
`means reveal a tendency towards an increase of band
`power synchronization which is most pronounced at oc-
`Ž
`.
`cipital recording sites during t2 cf. Fig. 4a . Hemispheric
`Ž
`.
`differences showing a general tendency of an increase in
`band power in the left hemisphere for all recording sites
`can be observed during t2 and t3.
`As in Expt. 1,
`the only significant finding for the
`lower-1 alpha band is a main effect for Time. Again, the
`respective means as depicted in Fig. 2 show a gradual
`increase in desynchronization from t1 to t3. The Scheffe´
`test for pairwise comparisons yields a critical difference of
`12.5 at the 5% level cf. Fig. 2b .
`Ž
`.
`
`Fig. 2. Event-related desynchronization ERD , reflecting task-related
`.
`Ž
`band power changes in the three experiments and the four frequency
`bands. A significant increase in desynchronization in one of the poststim-
`ulus intervals t2–t5 with respect to the prestimulus level of activity t1 is
`marked by an asterisk. Semantic memory demands are highest in Expt. 1
`but during t4 and t5 only. Note that only the upper alpha band responds
`selectively to this type of task demands.
`
`2-way interactions are included in the 3-way interaction,
`the description of this result which is depicted in Fig. 3
`will suffice to explain all of the remaining significant
`findings. Fig. 3 shows a pronounced increase in desyn-
`chronization which is much stronger over the left hemi-
`Ž
`sphere for frontal, central, temporal, and parietal record-
`.
`ing sites that can be observed primarily during t5 i.e.,
`Ž
`during the processing of the semantic task . Most interest-
`.
`ingly, the increase in desynchronization from t1 to t5 is
`significant for all of the five recording locations over the
`left hemisphere but for none of the right hemisphere. In the
`right hemisphere, a significant increase was found only
`from t1 to t4 cf. the respective arrows in Fig. 3 . The
`Ž
`.
`judgment of a significant increase from t1 to one of the
`poststimulus interval