Friedemann Pulvermueller (1) (1994) Brain Rhythms, Cell Assemblies and Cognition:. Psycoloquy: 5(48) Brain Rhythms (1)

Volume: 5 (next, prev) Issue: 48 (next, prev) Article: 1 (next prev first) Alternate versions: ASCII Summary
PSYCOLOQUY (ISSN 1055-0143) is sponsored by the American Psychological Association (APA).
Psycoloquy 5(48): Brain Rhythms, Cell Assemblies and Cognition:

Target Article by Pulvermueller et al. on Brain-Rhythms

Friedemann Pulvermueller (1)
Hubert Preissl (1)
Carsten Eulitz (2)
Christo Pantev (2)
Werner Lutzenberger (1)
Thomas Elbert (2)
Niels Birbaumer (1, 3)

(1) Institut fuer Medizinische Psychologie und
Verhaltensneurobiologie, Universitaet Tuebingen,
Gartenstrasse 29, 72074 Tuebingen, Germany

(2) Institut fuer Experimentelle Audiologie,
Universitaet Muenster, Kardinal von Galen-Ring 10,
48149 Muenster, Germany

(3) Universita degli Studi, Padova, Italy


In modern brain theory, cortical cell assemblies are assumed to form the basis of higher brain functions such as form and word processing. When gestures or words are produced and perceived repeatedly by the infant, cell assemblies develop which represent these building blocks of cognitive processing. This leads to an obvious prediction: cell assembly activation ("ignition") should take place upon presentation of items relevant for cognition (e.g., words, such as "moon"), whereas no ignition should occur with meaningless items (e.g., pseudowords, such as "noom"). Cell assembly activity may be reflected by high-frequency brain responses, such as synchronous oscillations or rhythmic spatiotemporal activity patterns in which large numbers of neurons participate. In recent MEG and EEG experiments, differential gamma-band responses of the human brain were observed upon presentation of words and pseudowords. These findings are consistent with the view that fast coherent and rhythmic activation of large neuronal assemblies takes place with word but not pseudowords.


brain theory, cell assembly, cognition, event related potentials (ERP), electroencephalograph (EEG), gamma band, Hebb, language, lexical processing, magnetoencephalography (MEG), psychophysiology, periodicity, power spectral analysis, synchrony


1. According to one theoretical view, higher brain functions are based on processing units called cell assemblies. Cell assemblies are large groups of cortical pyramidal neurons with strong reciprocal internal connections. Cell assemblies develop in a randomly connected neuronal network when sets of neurons are frequently active simultaneously so that their connections strengthen (Hebb's law, Hebb, 1949; Gustafsson et al., 1987; Bonhoeffer et al., 1989; Ahissar et al., 1992). Neurons making up one assembly do not need to be located in a small cortical area. They may be distributed widely over various cortical regions. Such TRANSCORTICAL ASSEMBLIES are likely to be held together through long axons of pyramidal cells known to connect distant cortical areas (Pandya and Yeterian, 1985; Braitenberg and Schuez, 1991; Deacon, 1992a; Deacon, 1992b). Because of strong intra-assembly connections, the activation of some neurons of an assembly leads to spreading activation in the network and, finally, to an IGNITION of the whole assembly [NOTE 1]. Overshooting activity will occur in the cortex when too many assemblies ignite at the same time. A regulatory mechanism is accordingly required to keep the level of cortical excitation close to a target value (Braitenberg, 1978). This regulatory mechanism guarantees that only one or a limited number of assemblies will ignite at a time.

2. What is the purpose of transcortical assemblies? One obvious answer is the following: they make it possible to "bind" information represented in different parts of the cortex. Such binding may occur between neurons in the visual system stimulated by features of a perceived object. If assemblies include various local clusters of neurons, they may well represent entities composed of numerous features, such as complex forms. Binding in widely distributed cell assemblies may also be relevant for associations between sensory modalities. Such associations are necessary for the representation of objects that can be perceived through various sensory channels. Transcortical assemblies may also be the neural counterpart of sensorimotor associations, a prerequisite for language acquisition. For example, in order to repeat syllables and words, the child needs to associate the auditory pattern perceived with a motor (articulatory) pattern. Information represented in distant parts of the cortex has to be bound in order to perform complex tasks such as form perception, object recognition, using and understanding language, and reasoning. These higher, or cognitive, brain capacities appear to require devices that bridge distances in the cortex and allow for fast between-area exchange of information. Transcortical assemblies are candidate mechanisms for subserving this function.

3. If Hebb's rule holds true, cell assemblies are formed during ontogenesis. This can be illustrated using language representation as a paradigm (Braitenberg, 1980; Braitenberg and Pulvermueller, 1992). During early language development, the infant frequently perceives sounds which lead to an "imprinting" of phonetic perception (Kuhl et al., 1992). Language sounds cause joint activation of neurons responding to their features; neurons jointly activated by the same phoneme develop into an assembly, the phoneme's neural correlate. In a later stage of language development, the infant repeatedly articulates sound sequences and word forms. These articulations are caused by activity in the motor cortex. Articulations cause acoustic signals which are fed back to the auditory cortex, where they lead to additional neuronal activation. Thus, during early articulations specific patterns of activity are present almost simultaneously in distant cortical regions: the motor and auditory systems. Because there are strong cortico-cortical projections connecting these systems, simultaneously activated neurons will strengthen their connections and develop into an assembly. This suggests that early articulations trigger the formation of transcortical assemblies corresponding to specific syllables and word forms. Note that such cell assemblies must comprise neurons of distant cortical regions in both hemispheres. It is well-known, however, that language is normally lateralized to the left hemisphere in most right-handers, suggesting that most neurons in language assemblies are located in the left hemisphere. If transcortical assemblies are the neural correlates of individual words, connections between such assemblies may be the physical realization of rules determining word sequences. In this case, the hierarchy of linguistic structures (phoneme, morpheme, word, sentence) has its biological equivalent in a hierarchy of cell assemblies corresponding to these cognitive entities (Braitenberg and Pulvermueller, 1992; Pulvermueller, 1992) [NOTE 2].

4. It must be emphasized that the transcortical assembly hypothesis is still somewhat speculative. However, a large body of evidence from various fields can be accounted for on the basis of the cell assembly approach. For example, the cell assembly theory of language processing accounts for data from aphasia research (Pulvermueller and Preissl, 1991; Pulvermueller, 1992; Pulvermueller and Schoenle, 1993; Pulvermueller and Preissl, 1994), from psycholinguistic experiments (Pulvermueller et al., 1993; Mohr et al., 1994), and from electrophysiological investigations of language processes (Lutzenberger et al., 1994; Pulvermueller et al., 1994a; Pulvermueller et al., 1994b). Hence, the concept of cell assemblies, though speculative, is useful.

5. The major advantage of a theory bridging neurobiology and cognition is the following: it allows for predictions which cannot be formulated on the basis of biological or cognitive theories alone. In this target article, we discuss a prediction of the cell assembly theory and some recent data obtained to test it.


6. If a cell assembly has strong internal connections, its ignition will take place instantaneously so that all or at least many of its neurons become active almost at the same time. The question of how activity will spread through the assembly has at least two possible answers. After activation of a subset of the assembly neurons, almost all its remaining neurons may be activated synchronously some time increment later. Figure 1a presents a sketch of such an assembly. If all neurons in this network have a threshold of 2 (i.e., they need 2 excitatory inputs to become active), simultaneous activation of the three neurons on the left will lead to synchronous activity of all assembly neurons a few time increments later. However, a second possibility is that the internal connections of a neural network may also determine a "stepwise" mode of activation. In the network schematized in Figure 1b, activity spreads from one row to the next after all neurons of the uppermost row have been activated. In this case, the activation process does not lead to synchronous activation of all assembly neurons but to a well-ordered spatiotemporal pattern of activity within the network. Hence only few assembly neurons become active at exactly the same time. The more likely case is that two neurons become active one after the other with a fixed delay. Although these simple examples are far from providing an adequate picture of cortical mechanisms, they can illustrate an important point. The activation of strongly coupled sets of neurons may take place synchronously or stepwise, according to a well-ordered spatiotemporal pattern of activity. The architecture of the network determines the activation process [NOTE 3].


               o. . . . . . . . .o                                  
              ...               ...                                 
            .  .  .           .  .  .                               
          .    .    .       .    .    .                             
        .      .      .   .      .      .                           
      o        .        .        .        o                         
        .      .      .   .      .      .                           
          .    .    .       .    .    .                             
            .  .  .           .  .  .                               
              ...               ...                                 
               o. . . . . . . . .o                                  

    FIGURE 1a: Simple model of a reciprocally connected cell assembly.
    Circles represent neurons, dots represent BIDIRECTIONAL
    connections between neurons. All neurons have a threshold of 2,
    i.e., they need two simultaneous inputs to become active. If the
    three neurons on the left are active, the rest of the assembly will
    also become active a little later.


               o       o       o                                    
               ..     ...      .                                    
               .  . .  .  .    .                                    
               .  . .  .    .  .                                    
               ..     ..      ..                                    
               o       o       o                                    
               .      ...     ..                                    
               .    .  .  . .  .                                    
               .  .    .  . .  .                                    
               ..      ..     ..                                    
               o       o       o                                    

    FIGURE 1b: Simple model of a neuron ensemble generating a fixed 
    spatiotemporal activity pattern. Circles represent neurons, 
    dots represent UNIDIRECTIONAL connections from upper to lower 
    neurons. All neurons have a threshold of 2. If the uppermost 
    three neurons are active, a wave of excitation will run through 
    the network.

7. If cell assemblies have strong and reciprocal internal connections, these connections warrant that activity is retained for some time within an activated assembly. Using terminology proposed by Braitenberg (1978), the assembly "holds" after its activation. The processes taking place while activity is retained may vary as a function of assembly architecture. For example, the network depicted in Figure 1a stays active; that is, all of its neurons will remain excited after their simultaneous activation. The network illustrated in Figure 2a (which is taken from Palm, 1982) shows a simple periodic activity pattern. If only two of the four neurons are active, the other two will become excited at the next time increment. (Again, neurons are assumed to have a threshold of 2.) One increment later, the original pair of neurons will again be excited, and so on. Each neuron of this network will repeatedly become active and inactive, that is, it will oscillate. A similar process takes place in the assembly illustrated in Figure 2b. Since these neurons form a "circle," activity will circulate within the network after one row of neurons has been activated. This results in repeated occurrences of the same spatiotemporal pattern. Again, each assembly neuron will show repetitive or periodic activity.


               o. . . . . . . . .o                                  
               ..               ..                                  
               .  .           .  .                                  
               .    .       .    .                                  
               .      .   .      .                                  
               .        .        .                                  
               .      .   .      .                                  
               .    .       .    .                                  
               .  .           .  .                                  
               ..               ..                                  
               o. . . . . . . . .o                                  

    FIGURE 2a: Illustration of an assembly generating oscillations.
    Circles represent neurons, dots represent BIDIRECTIONAL
    connections between neurons. All neurons have a threshold of 2. If
    two neurons of the assembly are active, the other two neurons will
    be active at the next time-increment, etc.


       . . . . . . . . .                                         
       .             . . .                                       
       .           .   .   .                                     
       .   . .   .     .     .   . .                             
       .   .   o       o       o   .                             
       .   .   ..     ...      .   .                             
       .   .   .  . .  .  .    .   .                             
       .   .   .  . .  .    .  .   .                             
       .   .   ..     ..      ..   .                             
       .   .   o       o       o   .                             
       .   .   .      ...     ..   .                             
       .   .   .    .  .  . .  .   .                             
       .   .   .  .    .  . .  .   .                             
       .   .   ..      ..     ..   .                             
       .   .   o       o       o   .                             
       .   . . .       .       . . .                             
       .               .                                         
       . . . . . . . . .                                         

    FIGURE 2b: Illustration of an assembly generating a more complex
    spatiotemporal activity pattern. Circles represent neurons, most
    dots represent UNIDIRECTIONAL connections from upper to lower
    neurons. Only the long connections (on the left and right) relay
    activity from bottom to top neurons. All neurons have a threshold
    of 2. After the uppermost three neurons are activated, a wave of
    excitation will circulate in the network.

8. Another mechanism generating periodic activity requires inhibitory connections (Wilson and Cowan, 1973). Consider the simple networks depicted in Figure 3a, where each neuron has an activation threshold of 1. The uppermost neuron receives constant excitatory input, so that it will be active initially. This leads to activation of the lower, inhibitory neuron, which will in turn lead to inhibition of the upper neuron. When the upper neuron is switched off by the inhibitory input, the inhibitor itself will become inactive one time-increment later. At this point, the constant input can again activate the upper neuron, which again inhibits itself through the inhibitory loop, resulting in oscillatory or periodic activity. If the excitatory neuron is interpreted not as a single neuron but as an assembly of neurons, such as the one depicted in Figure 1a, coherent and repetitive activation and deactivation of these neurons can be expected. Coherent oscillations will occur even if each excitatory neuron of such a network has its own inhibitor (and additional conditions are met; see Schuster and Wagner, 1990). If the uppermost excitatory neuron in Figure 3b is continuously activated, it will force its sister neurons into coherent oscillations after a short time lag.

9. Activation of a strongly coupled cell assembly leads to coherent activity of numerous neurons. Under certain conditions, this ignition process may lead to oscillatory activity (Fig. 2a) or to a more complex rhythmic activity pattern (Fig. 2b). If the assembly architecture itself does not determine periodic activity (see Fig. 1a), it is plausible that inhibitory regulatory processes will generate periodicity (Fig. 3b) [NOTE 4]. There is strong electrophysiological evidence that both processes -- coherent oscillations and patterned activity of neurons -- play an important role in cortical processing (Abeles, 1982; Gerstein et al., 1989; Abeles, 1991; Engel et al., 1992). Thus, it can be assumed that the ignition of an assembly includes periodic activity of a large group of neurons [NOTE 5]. During the ignition of a transcortical assembly, periodic activity should occur in various cortical areas.


           . . . o                                         
           .     T                                         
           .     .                                         
           .     .                                         
           .     .                                         
           .     .                                         
           .     .                                         
           .     .                                         
           .     o                                         
           .     .                                         
           . . . .                                         

    FIGURE 3a: Sketch of an oscillating circuit including an inhibitory
    element. All neurons have a threshold of 1. Connections ending in
    a  or a T are excitatory or inhibitory, respectively. If the upper
    neuron receives continuous input, both neurons will become active
    and inactive periodically.


           . . . o . . . . . . . . . . . .                 
           .     T .                     .                 
           .     .   .                   .                 
           .     .     .                 .                 
           .     .       .               .                 
           .     .         .             .                 
           .     .     . . . o     . . . o                 
           .     .     .     T     .     T                 
           .     o     .     .     .     .                 
           .          .     .     .     .                 
           .     .     .     .     .     .                 
           . . . .     .     .     .     .                 
                       .     .     .     .                 
                       .     .     .     .                 
                       .     o     .     o                 
                       .          .                      
                       .     .     .     .                 
                       . . . .     . . . .                 

    FIGURE 3b: Sketch of three coupled oscillators. All neurons have a
    threshold of 1. Connections terminating in a  or a T are
    excitatory or inhibitory, respectively. Undirected dotted lines are
    excitatory. If the uppermost neuron receives constant excitatory
    input, oscillations of all three neuron pairs will occur with a
    short phase lag.


10. If cell assemblies are the basic units of cognition, neuronal populations must become active when cognitive processes take place. This implies correlated activity of numerous neurons. Correlated post-synaptic activity in apical dendrites of numerous pyramidal cells will also affect the local field potential and, if the number of neurons participating in the pattern is large enough, the weighted sum of post-synaptic currents will lead to a surface potential which is visible in the EEG (Speckmann and Elger, 1982; Mitzdorf, 1985; Birbaumer et al., 1990). Magnetic fields caused by the intracellular currents flowing from the dendritic tree towards the soma can also be picked up in the MEG if pyramidal cells are located in sulci, thus causing currents to flow tangentially to the surface of the head (Cuffin and Cohen, 1979; Hari and Lounasmaa, 1989). Ignition of a cell assembly implies fast activation of a large neuronal population. This predicts a sharp rise in EEG and MEG. After activation, an assembly with reciprocal connections will retain its activity and correlated periodic firing of assembly neurons takes place. In this case, EEG and MEG peaks may occur repeatedly [NOTE 6]; hence the EEG and MEG signals will include enhanced spectral power in at least one frequency band. If, for example, the ignition process includes synchronous 40-Hz oscillations of assembly neurons, the spectral power in the 40-Hz range should be specifically affected. If ignition consists of repetitive spatiotemporal activity patterns, a more complex change results which may affect several frequency bands. Such a complex change will nevertheless also be visible in specific frequencies. On the basis of theoretical considerations it is hardly possible to determine the exact frequency range which will be affected by assembly ignition. Cell assembly ignition will most probably cause electrophysiological changes visible in the evoked potential as well as in low and high frequency bands (Elbert, 1993), but there is at least one neuroanatomical reason for assuming that rhythmic activity patterns can be seen in high frequency bands. It is well known that there are numerous cortico-cortical axons with conduction velocities around 10 m/s and faster (Aboitiz et al., 1992). In such axons, action potentials travel from Broca's to Wernicke's region within 10-20 ms (Lines et al., 1984; Saron and Davidson, 1989), so that the "round trip time" from Wernicke to Broca and back to Wernicke would be about 20-40 ms. It is therefore likely that the spread of activity and reverberation within an assembly will take place quickly, that is, in the range of several milliseconds. This suggests that changes in spectral power will be seen in the gamma band (20 Hz and up).

11. According to cell assembly theory, ignition will take place when an assembly is stimulated appropriately. Thus, the theory predicts that cell assembly ignition takes place in the cortex after an adequate stimulus has been perceived. In this case, large numbers of neurons become active simultaneously and may join a periodic pattern thereafter. This might be visible in the gamma-band response of the cortex. If cognitive processing does not take place, cell assemblies do not become active. More precisely, cell assemblies may be stimulated and, therefore, become slightly active, but no full activation, no ignition, should take place. Gamma-band responses should therefore be reduced. In summary, gamma-band spectral power should be strong when cognitive processing takes place but reduced when such processing does not occur.


12. Gamma-band responses have been observed in various mammals (including humans) using different methods, such as single and multiple unit recordings or local field potential, ECoG, EEG and MEG recordings. Gamma-band activity can be elicited by viewing moving bars (Eckhorn et al., 1988; Engel et al., 1992), by simple auditory stimuli such as tones (Pantev et al., 1991), by somatosensory stimuli (Ahissar and Vaadia, 1990) and by odors (Bressler and Freeman, 1980; Freeman, 1991). They also accompany manipulative movements (Murthy and Fetz, 1992). However, this does not imply a specifically cognitive function for gamma-band responses. Very simple stimuli, such as bars or tones, are unlikely to elicit cognitive processing. Such stimulation triggers perception processes, but it is unclear whether additional processes will follow.

13. To determine whether synchronized and/or repetitive activity of neurons serves a specifically cognitive function, one must compare brain responses in two paradigms only one of which induces cognitive processing of a certain kind. In recent experiments it was found that two bars moving in the same direction lead to synchronous and fast oscillatory activity of neurons activated by the stimuli (Eckhorn et al., 1988; Gray and Singer, 1989). These neurons may be located in distant cortical areas. In contrast, if two bars move in different directions, two neurons, each responding to one of the stimuli, fail to respond synchronously (Gray et al., 1989; Engel et al., 1990). Although this indicates that cortical spatiotemporal responses change with Gestalt features (Engel et al., 1992; Singer, 1994), such as continuity, it is still unclear how these responses relate to perception of complex forms. A complex moving stimulus may well lead to the perception of lines moving in different directions. If cortical synchrony was an indicator of Gestalt integration, it could instead be expected that two moving stimuli invoke more complex Gestalt integration processes and, therefore, more pronounced synchronized activity (von der Malsburg, p.c.). It may be argued, however, that the two lines will not be integrated into a unitary form. For this experiment, it is difficult to decide which cognitive processes are triggered by the differing stimuli.

14. More and less complex manipulative movements are also correlated with different patterns of synchronized oscillatory brain activity in the gamma range. When a monkey performs a complex movement, such as retrieving raisins from the slots of a Kluever board, synchronized gamma-band activity can be recorded from the motor and somatosensory cortices (Murthy and Fetz, 1992). Coherent gamma-band activity was reduced when monkeys performed simple movements, such as alternating flexion and extension of the wrist. These results were confirmed, in part, by an MEG investigation of human brain responses during complex movements (Kristeva-Feige et al., 1993). During a complex manual task, enhanced gamma-band activity around 30 Hz was found. The enhancement of gamma-band responses around 30 Hz might be attributable to the level of attention or the amount of sensorimotor integration required by complex movements. However, the complexity of the muscular activity could also be critical for the gamma-band response to occur. Although these results are consistent with the assumption that cognitive processes, that is, selective attention to sensorimotor integration, underlie stronger gamma-band responses, they might also arise from the complexity of the movement to be performed.

15. Another investigation of MEG responses of the human brain indicates that gamma-band responses reflect cognitive processing. Llinas and Ribary (1993) found reduced 40-Hz activity during delta sleep, whereas oscillatory activity was stronger during both wakefulness and REM sleep. Because vivid dreaming (which usually occurs during REM sleep) and wakefulness imply cognitive processing, the authors propose that enhanced gamma-band responses are a correlate of cognition. It can be argued, however, that various other variables (arousal level, brain activity level, etc.) distinguish delta sleep from REM sleep or wakefulness.

16. In earlier studies, EEG recordings were used to investigate changes of gamma-band activity associated with tasks, such as verbal and visual-spatial problem solving. Sheer and coworkers (Spydel et al., 1979; Sheer, 1984) reported increased spectral power around 40 Hz when subjects engaged in cognitive tasks requiring the focussing of attention. A possible methodological problem in these studies, however, is EEG artifact caused by muscle activity. Although these authors report no correlation between EEG and electromyographic (EMG) spectral responses recorded from temporal and splenius muscles, one may well argue that changes in 40-Hz power can be caused by muscles not monitored in these experiments. This problem can be solved by investigating spectral responses of even higher frequencies. Spectral power of EMG responses increases with frequency until at least 80 Hz (Cacioppo et al., 1990). Thus, if muscle activity causes differences, for example, in the 40-Hz range, the same effects must be present (and must be even more pronounced) in higher frequency intervals, for example, around 60 or 80 Hz. If no differences occur in these bands, muscle activity cannot be the cause of an effect visiblt in lower frequencies.

17. To determine whether cognitive processing implies specific changes in high frequency brain responses, two conditions must be compared that only differ with regard to the cognitive processes they elicit. Comparing responses to bars and "meaningful" pictures of objects would be one option. However, bars and pictures have very different perceptual complexity and a difference in evoked spectral responses could be a consequence of this physical difference. Two stimulus classes that are of equal perceptual complexity but trigger distinct cognitive processes are words and meaningless pseudowords made up of the same letters. A word, such as "moon", will immediately be accessed and comprehended, whereas a matched pseudoword, such as "noom," will fail to elicit immediate lexical access and comprehension. On the cortical level, words should lead to cell assembly ignition, whereas pseudowords should fail to ignite specific assemblies. This leads to an obvious prediction: gamma-band responses to words should be stronger than responses to pseudowords.


18. The following experiment was carried out to investigate EEG responses to words and pseudowords. Fifteen right-handed native German speakers performed lexical decisions on visually presented words and matched pseudowords (64 stimuli of each category; some examples: arbeit, glaube, naemlich vs. teibra, gleuba, maelinch). Each word was presented for 100 ms in the center of a video screen. The interstimulus interval varied between 3.5-4.5 s. Words subtended 0.5 degrees of vertical visual angle and a maximum of 2.3 degrees of horizontal visual angle. They were written in black letters on a grey background. Lexical decisions were expressed by moving the index finger of the left hand. Word/pseudoword decisions were expressed by moving a switch to left or right (switch-directions and word/pseudoword decisions were counterbalanced between subjects). EEG was recorded through 17 tin electrodes against linked mastoids. Six of the electrodes were placed close to the perisylvian regions of each hemisphere, respectively [NOTE 7]. The EEG was recorded for 1.28 s per trial, starting 0.1 s before stimulus presentation (0.1 s baseline). Event-related potentials (ERPs) were calculated for each electrode, condition, and subject (artifact rejection and correction as described in Elbert et al., 1985). To obtain reference-free data, artifact-free signals obtained from single trials were also submitted to current source density (CSD) analyses. CSD data were filtered in three frequency bands: 25-35 Hz, 35-45 Hz, and 55-65 Hz. These data were normalized by dividing them by the average value obtained in the respective baseline. Finally, mean normalized evoked spectral power was calculated for both conditions (words/pseudowords), for each subject, and for each electrode. Average values in three time windows were evaluated: 120-320 ms, 320-520 ms and 520-720 ms after stimulus onset. Three-way analyses of variance were carried out (design: Wordness (word/pseudoword) x Hemisphere (left/right) x Site (six electrodes from anterior to posterior)). Greenhouse-Geisser correction of degrees of freedom was applied when adequate. For a more detailed description of the methods, see Lutzenberger, Pulvermueller and Birbaumer (1994).

19. Figure 4 displays ERPs evoked by word and pseudoword presentation. Around 400 ms post-stimulus onset, pseudowords elicited more negative ERPs compared to words. There was a significant main effect of Wordness in the 320-520 ms time window, F (1,14) = 12.6, p < 0.003. A larger late negativity after pseudowords compared to words had been previously reported (Holcomb and Neville, 1990). This larger negativity can be taken as evidence that pseudoword presentation elicited more activity, that is, a larger number of excitatory post-synaptic potentials, in apical dendrites of numerous neurons (Speckmann and Elger, 1982; Mitzdorf, 1985; Birbaumer et al., 1990).


  U   |                               x x x x  words         
[uV] |
  -.4 |                               o o o o  pseudowords   
  -.2 |                                                      
      |         x x                                          
    0 |        x   x                                         
      | x     x     x                 o o o o                
  +.2 |   x  x       x              o          o  o          
      |     x         x            o     x           o o     
  +.4 |                x          o    x   x x x          o  
      |                 x        o   x           x  x  x  x  
  +.6 |                   x     o  x                         
      |                      x x x                           
  +.8 |                                                      
             100       200       300       400       500  [msec]

    FIGURE 4: Results from EEG experiment. Event-related potentials
    (ERPs) after word and pseudoword presentation. Data are averaged
    over 15 subjects. While there is no difference between word and
    pseudoword-evoked potentials up to 300 msec, a late negative-going
    component around 400 ms after stimulus onset is significantly
    LARGER after pseudowords.

20. Figure 5a presents normalized evoked spectral responses in the 30-Hz (25-35 Hz) band recorded from the left perisylvian cortices. Because statistical analysis did not reveal any reliable effect of the factor Site, averaged data from all six electrodes are displayed in this diagram. Spectral responses to words and pseudowords differed around 400 ms after stimulus onset. Statistical analysis revealed a significant Wordness by Hemisphere interaction for the time interval between 320 and 520 ms, F (1,14) = 8.4, p < 0.01. This interaction is displayed in Figure 5b. Word presentation did not change 30-Hz power compared to the baseline. In contrast, pseudowords elicited reduction of 30-Hz cortical responses over the left hemisphere. This deflection was small (1-2 percent of the baseline values). However, it was consistently observed in most subjects tested (14 out of 15). In contrast, analysis of all other frequency bands did not reveal a similar interaction in any of the time windows.


normalized power 25-35 Hz [arbitrary units] x x x x words

   1. |                               o o o o  pseudowords   
   .5 |                                                      
      |                                    x  x              
    0 |x  x                 x  x  x  x  x        x        
      |   o  x  x  x  x  x                          x  x  x  x
  -.5 |      o                                                 o
      |        o  o                                         o
  -1. |              o                                   o   
      |                 o        o                    o      
 -1.5 |                    o  o     o     o  o  o  o         
      |                                o                     
  -2. |                                                      
             100       200       300       400       500  [msec] 

    FIGURE 5a: Results from EEG experiment. Normalized evoked spectral
    power (arbitrary units) between 25 and 35 Hz recorded over the left
    hemisphere. Spectral power elicited by words and pseudowords is
    plotted against time after stimulus presentation. Responses are
    averaged over 15 subjects. Around 400 ms post stimulus onset, 30-Hz
    activity is REDUCED over the left hemisphere after pseudowords
    ("pseudoword depression"). There was no such difference over the
    right hemisphere.


normalized power 25-35 Hz [arbitrary units]

   1. |                                     |          
      |        |                            |          
   .5 |        |                            |          
      |        |                            O          
    0 |        X                            O          
      |              O                X                
  -.5 |              O                |                
      |              O                |                
  -1. |              O                |                
      |              O                                 
 -1.5 |              O                           XXX words
      |              O                                 
  -2. |              |                           OOO pseudo
      |              |                               words
      |              |                                 
                left                  right            
             hemisphere             hemisphere         

    FIGURE 5b: The significant Hemisphere x Wordness interaction is
    displayed. Averaged normalized 30-Hz power in the interval between
    320 and 520 msec recorded from the left and right hemisphere. Lines
    indicate standard errors.

21. This difference cannot be the result of an artifact caused by muscle activity. As noted earlier, the power spectrum of potential changes caused by muscle contractions increases monotonically with frequency until around 80 Hz (Cacioppo et al., 1990). Thus, a muscle artifact affecting the 30-Hz band must also be visible in the 60-Hz range. The fact that no significant interactions or main effects were observed in the highest frequency band examined (60-Hz band) indicates that muscle potentials did not cause the effect (for further discussion, see Lutzenberger et al., 1994).

22. It may be argued that the reduction of gamma-band power after pseudowords was the consequence of some other recording or evaluation artifact. At this point, it also cannot be excluded that this reduction was related to the language used in the experiment (German), to visual stimulus presentation, or to the motor response the participants had to perform. We therefore carried out another experiment in which all these features of experimental setting and evaluation procedure were changed. Biomagnetic signals were recorded from both hemispheres of five right-handed native speakers of English who heard English words and pseudowords spoken by a professional speaker. In this case, 30 items of each stimulus category (words/pseudowords) were repeated four times, resulting in 120 tokens of each category (examples: moon, room, arm vs. noom, mool, arn). The interstimulus interval varied between 2.5 - 3.5 s. Subjects did not have to respond to the stimuli. However, they were asked to memorize all stimuli in order to answer questionnaires which were posed during breaks. In this case, no CSD analysis was necessary, since MEG provides reference-free data. To evaluate spectral responses a method described by Makeig (1993) was used. For a variety of frequency bands (width: 9.6 Hz), spectral power was calculated in overlapping time windows of 0.3 s. Two adjacent time windows overlapped by 50 percent. Spectral power values were, again, normalized (0.6 s baseline), and averaged. (For further details, see Pulvermueller et al., 1994).

23. Figure 6 presents normalized spectral responses of one subject to words and pseudowords obtained from one channel over the left hemisphere. While no pronounced change was elicited by presentation of words, a marked deflection of spectral power around 30 Hz followed pseudoword presentation. Reduction of 30-Hz spectral power after pseudowords was observed over the left hemisphere of all five subjects tested. No consistent change in spectral power was present in any of the other frequency bands, or over the right hemisphere. These changes were only seen at anterior channels located above the left inferior frontal lobe. In these channels, magnetic fields evoked by words and pseudowords were larger compared to all other channels; hence maximal signal-to-noise-ratio can be assumed.


normalized power 24-33 Hz [% of baseline] x x x x words

  116 |                               o o o o  pseudowords    
  108 |                                                       
  100 | x  x  x                          x  x  x  x     x  x  
      |       o  x           x  x  x  x              x        
   92 |             x  x  x                                   
      |          o                                           o
   84 |            o        o                          o  o   
      |               o  o     o  o     o  o  o  o  o         
   76 |                              o                        
             100       200       300       400       500  [msec]

    FIGURE 6: Results from MEG experiment. Normalized spectral power
    between 24 and 33 Hz evoked by words and pseudowords. Data were
    recorded from one channel over the left hemisphere of one
    individual. After pseudoword presentation, gamma-band activity was
    REDUCED relative to the baseline. No such difference was present
    over the right hemisphere. The same pattern of results was obtained
    for all 5 right-handed individuals tested.

24. The consistency of the results of the EEG and the MEG experiments makes it unlikely that pseudoword-specific gamma-depression is an artifact caused by recording or quantification procedures or that it is affected by parameters such as experimental language, stimulus modality, or response mode [NOTE 8].



25. Based on these data, the original prediction of stronger gamma-band responses to words compared to pseudowords can obviously be supported. A preliminary explanation of this difference is the following: cell assembly ignition takes place after word presentation. This leads to correlated and repetitive activation of large neuronal sets which can be observed in EEG and MEG responses. In contrast, pseudoword presentation does not lead to ignition. Hence 30-Hz power is reduced after pseudowords. Activity differences in the gamma band are primarily seen over the left hemisphere, because most neurons of language assemblies are located in the left hemisphere.

26. This preliminary hypothesis can be contested for several reasons. First, the cell assembly theory would predict enhanced gamma-band activity after words compared to the baseline, rather than a reduction after pseudowords. However, it can be assumed that word processing does not take place only immediately after stimulus presentation. Word processing and thinking about these words may last for a few seconds, that is, throughout the entire interstimulus interval (which was 2.5-3.5 s in the MEG experiment and 3.5-4.5 s in the EEG experiment). This is consistent with the cell assembly theory. It was postulated that each activated assembly would "hold" for some time. Word-evoked spectral responses may therefore have contaminated the baseline and pseudoword-specific gamma-depression may be interpreted as a power decrease relative to word processing. However, it is not evident that subjects still processed words during baseline intervals. One can argue that subjects may have engaged in any kind of cognitive processing. This suggests that the normal (baseline) state of the brain is a continuous stream of thoughts, that is, a continuous succession of assembly ignitions. Only if one perceives very unusual material that cannot be processed immediately is there a temporary ignition failure.

27. A second possible objection against the reported results concerns the size of the difference between responses to words and pseudowords. These differences are rather small in the EEG experiment (about 1-2 percent of the baseline power) and even the most pronounced differences seen in MEG recordings only amount to some 20 percent of the baseline. This indicates that the differences induced by stimulus presentation were small compared to cortical background activity. This is not surprising, however. Probably not all cortical neurons participate in processing information contained in a stimulus word. Perhaps only one cell assembly is activated after a word has been presented. Assemblies are usually assumed to comprise 106 neurons or fewer (Palm, 1993), that is, only a small portion of the 1010 to 1011 neurons of the cortex. From this perspective, it is rather surprising that a small percentage of neurons generates a change in global activity of one to several percent.

28. A third possible objection addresses the relationship between spectral responses and event-related potentials (ERPs). Pseudowords lead to larger late negativities around 400 ms after stimulus onset, whereas 30-Hz power is reduced in this interval compared to words. If both event-related potentials and spectral responses were an indicator of arousal, attention, and/or neuronal mobilization processes, there would be an obvious incompatibility. It appears that ERPs and gamma-band responses reflect different processes. A cognitive and neurobiological model has to account for both 30-Hz depression and larger late negativities evoked by pseudowords.

29. On the psychological level, 30-Hz activity and late negativities in these experiments may be indicators of lexical access, lexical processing or lexical search. When a word is perceived, comprehended, or stored, relatively strong 30-Hz activity is observed. The possible psychological processes underlying gamma depression may be comprehension failure or an unsuccessful lexical search. If no equivalent of a stimulus word is found in the "cortical lexicon," 30-Hz activity decreases. However, there may be an intense lexical search when pseudowords are perceived, whereas search processes after word presentation terminates much faster, since the stimulus can be matched to a mental representation. It has been found that more intense memory search processes induce larger negativities in the event-related potential (Rockstroh et al., 1989; Roesler et al., 1993). Thus, one can speculate that 30-Hz power and late event-related potentials indeed mirror distinct cognitive processes. Whereas 30-Hz power indicates whether lexical access and processing take place, late negativities indicate the intensity of lexical search processes. According to another view, the reported results are related to attention and arousal processes. Pseudowords have never been perceived and are therefore highly unexpected. Thus, they cause enhanced arousal, whereas words do not. In contrast, word presentation leads to processing of only one word and to focussing on that stimulus, whereas pseudoword presentation may activate various associations. According to the latter view, stronger gamma-band responses to words would correspond to the process of selectively attending to words and late negative evoked potentials would covary with general arousal processes. Postulating an a priori correspondence between cognitive processes and psychophysiological measures has not been considered very satisfactory, however, and has been dubbed "psychophysio-analytic" by some researchers. It would be desirable to specify neural mechanisms for such cognitive processes that also correspond to known neural responses. The present data are explained only in terms of neural processes.

30. The following neurobiological model can account for both ERP and 30 Hz differences: after word presentation, the ignition of exactly one assembly takes place (while competing assemblies are inhibited). This ignition causes a small negative shift in the ERP. Because assembly ignition implies fast, periodic, correlated activity in a large number of assembly neurons, strong 30-Hz responses are present. Processes taking place after pseudoword presentation are the following: not only one, but several assemblies are preactivated to some degree. However, no full activation, no ignition, takes place in any of the preactivated assemblies. These assemblies are the neuronal equivalents of words phonologically similar to the pseudoword stimulus (the stimulus "noom" may activate assemblies corresponding to words such as noon, moon, room, etc.). Because several assemblies are activated, the sum of cortical excitation is larger than in full ignition of only one assembly. This results in relatively large late negativities in the event-related potential. However, neuronal activity in several assemblies is not correlated, that is, activation-deactivation cycles differ between assemblies [NOTE 9]. In global recordings of cortical activity, the effects of fast uncorrelated and periodic activity tend to cancel one another. Reduced high-frequency responses are therefore observed at recording electrodes and channels. In psychological terms, the ignition of a word-specific assembly corresponds to lexical access and processing (which implies focussed attention to one particular word), and preactivation of several assemblies without ignition can be considered the correlate of an unsuccessful search process (which may be linked to enhanced global arousal).


NOTE 1. Note that stimulation will not necessarily lead to the full activation (ignition) of an assembly. Weak stimulation may lead to a small amount of (pre-)activity in an assembly, but not to full ignition. In contrast, strong stimulation of an assembly will cause an ignition.

NOTE 2. It has recently been shown that grammatical rules and even complex syntactic structures can be represented in a neural system comprising cell assemblies (Pulvermueller 1993, 1994a, 1994b).

NOTE 3. In this target article, cell assemblies are defined as sets of neurons with strong reciprocal connections (Braitenberg, 1978). Note that in a feedforward net such as the one depicted in Figure 1b connections are not reciprocal. Hence such a synfire chain (Abeles, 1982) will not be called an assembly here.

NOTE 4. Note that periodicity may be caused by cortical or subcortical mechanisms. The oscillator in Figure 3a could have its counterpart in excitatory and inhibitory neurons in the cortex. However, inhibition could also be provided by complex loops involving neurons in the basal ganglia and/or the thalamus.

NOTE 5. We note that this is not necessarily so. It may be that activation of a single assembly generates aperiodic or even chaotic activity patterns. At this point, however, we know of no strong evidence supporting this view. It will therefore be assumed that cell assembly ignition leads to periodic neural activity or, at least, to repetitive activation of large neuronal populations. In the latter case, the circulation time of activity in an assembly may vary within certain limits.

NOTE 6. It is possible that repetitive assembly activation will not be visible in the EEG or MEG. First, it may be too small to cause changes in these measures. Second, equal numbers of neurons may be active at each time-step and their activity may contribute equally to the recorded potential or field. Nevertheless, if numerous neurons are active at each time-step but contribute differently to the surface potential, it is likely that periodic activity will cause changes at the recording electrode.

NOTE 7. Electrode placement was as follows: To monitor activity over the perisylvian language cortices, nonstandard recording sites were chosen. P3 and P4 of the 10/20 system were moved one tenth of the nasion-inion distance away from Pz ("Angular electrodes"), and T3 and T4 were moved the same distance towards Cz ("Sylvian electrodes"). Furthermore, electrodes were placed half way between Sylvian electrodes and F7/F8 ("Broca electrodes") and half way between Sylvian and Angular electrodes ("Wernicke electrodes"). In addition, F7, F8, O1 and O2 of the 10/20 system were chosen. This resulted in two lateral lines of electrodes each close to one of the perisylvian regions.

NOTE 8. Another factor which does not appear relevant for the occurrence of the 30-Hz effect is the number of stimulus repetitions. The 30-Hz effect was not only observed after initial presentations of words and pseudowords; in another experiment, it was also present after several (up to 10) repetitions of the stimuli.

NOTE 9. Strictly speaking, this implies the following: if periodic activation is caused by inhibitory elements, these inhibitors must be assembly-specific. Because inhibitory neurons are small and therefore do not reach distant pyramidal cells, it can be assumed that each local cluster of assembly neurons activates only its own inhibitors.

Acknowledgments: We wish to thank our nine PSYCOLOQUY referees for their helpful comments. This research was supported by a grant of the Deutsche Forschungsgemeinschaft (DFG Pu 97/2) and by a Helmholtz Fellowship for Neurobiological Research of the Bundesministerium fuer Forschung und Technologie (both to the first author) and by DFG grant SFB 307/B1 (to the fifth and seventh authors).


Abeles, M. (1982) Local cortical circuits. An electrophysiological study. Berlin: Springer Verlag.

Abeles, M. (1991) Corticonics - Neural circuits of the cerebral cortex. Cambridge: Cambridge University Press.

Aboitiz, F., Scheibel, A.B., Fisher, R.S. & Zaidel, E. (1992) Fiber composition of the human corpus callosum. Brain Research 598:143-153.

Ahissar, E. & Vaadia, E. (1990) Oscillatory activity of single units in a somatosensory cortex of an awake monkey and their possible role in texture analysis. Proceedings of the National Academy of Sciences 87:8935-8939.

Ahissar, E., Vaadia, E., Ahissar, M., Bergman, H., Arieli, A. & Abeles, M. (1992) Dependence of cortical plasticity on correlated activity of single neurons and on behavior context. Science 257:1412- 1415.

Birbaumer, N., Elbert, T., Canavan, A.G.M. & Rockstroh, B. (1990) Slow Potentials of the Cerebral Cortex and Behavior. Physiological Reviews 70:1-41.

Bonhoeffer, T., Staiger, V. & Aertsen, A.M.H.J. (1989) Synaptic plasticity in rat hippocampal slice cultures: local "Hebbian" conjunction of pre- and post-synaptic stimulation leads to distributed synaptic enhancement. Proceedings of the National Academy of Sciences 86:8113-8117.

Braitenberg, V. (1978) Cell assemblies in the cerebral cortex. In Theoretical approaches to complex systems. (Lecture notes in biomathematics, vol. 21) R. Heim & G. Palm, eds. Berlin: Springer Verlag, pp. 171-188

Braitenberg, V. (1980) Alcune considerazione sui meccanismi cerebrali del linguaggio. In L'accostamento interdisciplinare allo studio del linguaggio. G. Braga, V. Braitenberg, C. Cipolli, E. Coseriu, S. Crespi-Reghizzi, J. Mehler & R. Titone, eds. Milano: Franco Angeli Editore, pp. 96-108

Braitenberg, V. & Pulvermueller, F. (1992) Entwurf einer neurologischen Theorie der Sprache. Naturwissenschaften 79:103-117.

Braitenberg, V. & Schuez, A. (1991) Anatomy of the cortex. Statistics and geometry. Berlin: Springer Verlag.

Bressler, S.L. & Freeman, W.J. (1980) Frequency analysis of olfactory system EEG in cat, rabbit and rat. Electroencephalography and Clinical Neurophysiology 50:19-24.

Cacioppo, J.T., Tassinary, L.G. & Fridlund, A.J. (1990) The skeletomotor system. In Principles of psychophysiology. Physical, social, and inferential elements. J.T. Cacioppo & L.G. Tassinary, eds. Cambridge: Cambridge University Press, pp. 325-384

Cuffin, B.N. & Cohen, D. (1979) Comparison of the magnetoencephalogram and the electroencephalogram. Electroencephalography and Clinical Neurophysiology 47:132-146.

Deacon, T.W. (1992a) Cortical connections of the inferior arcuate sulcus cortex in the macaque brain. Brain Research 573:8-26.

Deacon, T.W. (1992b) The neural circuitry underlying primate calls and human language. In Language origin: a multidisciplinary approach. J. Wind, B. Chiarelli, B.H. Bichakjian, A. Nocentini & A. Jonker, eds. Dordrecht: Kluwer Academic Publishers, pp. 121-162.

Eckhorn, R., Bauer, R., Jordan, W., Brosch, M., Kruse, W., Munk, M. & Reitboeck, H.J. (1988) Coherent oscillations: a mechanism of feature linking in the visual cortex? Multiple electrode and correlation analysis in the cat. Biological Cybernetics 60:121-130.

Elbert, T. (1993) Slow Cortical Potentials reflect the regulation of cortical excitability. In Slow potential changes in the human brain. V.C. McCallum & S.H. Curry, eds. New York: Plenum Press, pp. 235-251.

Elbert, T., Lutzenberger, W., Rockstroh, B. & Birbaumer, N. (1985) Removal of ocular artifacts from the EEG - a biophysical approach to the EOG. Electroencephalography and Clinical Neurophysiology 60:455- 463.

Engel, A.K., Koenig, P., Gray, C.M. & Singer, W. (1990) Stimulus- dependent neuronal oscillations in cat visual cortex: inter-columnar interaction as determined by cross-correlation analysis. European Journal of Neuroscience 2:588-606.

Engel, A.K., Koenig, P., Kreiter, A.K., Schillen, T.B. & Singer, W. (1992) Temporal coding in the visual cortex: new vistas on integration in the nervous system. Trends in Neurosciences 15:218-226.

Freeman, W.J. (1991) The Physiology of Perception. Scientific American February, 264:2, pp. 78-85.

Gerstein, G.L., Bedenbaugh, P. & Aertsen, A.M.H.J. (1989) Neuronal assemblies. IEEE Transactions on Biomedical Engineering 36:4-14.

Gray, C.M., Koenig, P., Engel, A.K. & Singer, W. (1989) Oscillatory responses in cat visual cortex exhibit inter-columnar synchronization which reflects global stimulus properties. Nature 338:334-337.

Gray, C.M. & Singer, W. (1989) Stimulus-specific neuronal oscillations in orientation columns of cat visual cortex. Proceedings of the National Academy of Sciences 86:1698-1702.

Gustafsson, B., Wigstroem, H., Abraham, W.C. & Huang, Y.Y. (1987) Long term potentiation in the hippocampus using depolarizing current pulses as the conditioning stimulus to single volley synaptic potentials. Journal of Neuroscience 7:774-780.

Hari, R. & Lounasmaa, O.V. (1989) Recording and interpretation of cerebral magnetic fields. Science 244:432-236.

Hebb, D.O. (1949) The organization of behavior. A neuropsychological theory. New York: John Wiley.

Holcomb, P.J. & Neville, H.J. (1990) Auditory and visual semantic priming in lexical decision: a comparison using event-related brain potentials. Language and Cognitive Processes 5:281-312.

Kristeva-Feige, R., Feige, B., Makeig, S., Ross, B. & Elbert, T. (1993) Oscillatory brain activity during human sensorimotor integration. NeuroReport 4:1291-1294.

Kuhl, P.K., Williams, K.A., Lacerda, F., Stevens, K.N. & Lindblom, B. (1992) Linguistic experience alters phonetic perception in infants by 6 months of age. Science 255:606-608.

Lines, C.R., Rugg, M.D. & Milner, A.D. (1984) The effect of stimulus intensity on visual evoked potential estimates of interhemispheric transmission time. Experimental Brain Research 57:89-98.

Llinas, R. & Ribary, U. (1993) Coherent 40-Hz oscillation characterizes dream state in humans. Proceedings of the National Academy of Sciences 90:2078-2081.

Lutzenberger, W., Pulvermueller, F. & Birbaumer, N. (1994) Words and pseudowords elicit distinct patterns of 30-Hz EEG responses in humans. Neuroscience Letters (in press).

Makeig, S. (1993) Auditory event-related dynamics of the EEG spectrum and effects of exposure to tones. Electroencephalography and Clinical Neurophysiology 86:283-293.

Mitzdorf, U. (1985) Current source density method and application in cat cerebral cortex: investigation of evoked potentials and EEG phenomena. Physiological Reviews 65:37-100.

Mohr, B., Pulvermueller, F. & Zaidel, E. (1994) Lexical decision after left, right and bilateral presentation of content words, function words and non-words: evidence for interhemispheric interaction. Neuropsychologia 32:105-124.

Murthy, V.N. & Fetz, E.E. (1992) Coherent 25- to 35-Hz oscillations in the sensorimotor cortex of awake behaving monkeys. Proceedings of the National Academy of Sciences 89:5670-5674.

Palm, G. (1982) Neural assemblies. Berlin: Springer Verlag.

Palm, G. (1993) On the internal structure of cell assemblies. In Brain theory: spatio-temporal aspects of brain function. A. Aertsen, ed. Amsterdam: Elsevier, pp. 261-270.

Pandya, D.N. & Yeterian, E.H. (1985) Architecture and connections of cortical association areas. In Cerebral cortex. Vol. 4. Association and auditory cortices. A. Peters & E.G. Jones, eds. London: Plenum Press, pp. 3-61.

Pantev, C., Makeig, S., Hoke, M., Galambos, R., Hampson, S. & Gallen, C. (1991) Human auditory evoked gamma-band magnetic fields. Proceedings of the National Academy of Sciences 88:8996-9000.

Pulvermueller, F. (1992) Constituents of a neurological theory of language. Concepts in Neuroscience 3:157-200.

Pulvermueller, F. (1993) On connecting syntax and the brain. In Brain theory - spatio-temporal aspects of brain function. A. Aertsen, ed. New York: Elsevier, pp. 131-145.

Pulvermueller, F. (1994a) Syntax und Hirnmechanismen. Perspektiven einer multidisziplinaeren Sprachwissenschaft. Kongitionswissenschaft 4:17-31.

Pulvermueller, F. (1994b) What neurobiology can buy language theory. Studies in Second Language Acquisition (in press).

Pulvermueller, F., Eulitz, C., Pantev, C., Mohr, B., Feige, B., Lutzenberger, W., Elbert, T. & Birbaumer, N. (1994) Gamma band brain responses reflect cognitive processing: an MEG study (submitted).

Pulvermueller, F., Lutzenberger, W. & Birbaumer, N. (1994) Electrocortical distinction of vocabulary types. Electroencephalography and Clinical Neurophysiology (in press).

Pulvermueller, F., Mohr, B., Rayman, J., Zaidel, E. & Aertsen, A. (1993) Bilateral presentation of words facilitates lexical processing in normals but not in split-brain patients. Society for Neuroscience Abstracts 19:1808.

Pulvermueller, F. & Preissl, H. (1991) A cell assembly model of language. Network 2:455-468.

Pulvermueller, F. & Preissl, H. (1994) Explaining aphasias in neuronal terms. Journal of Neurolinguistics 8:75-81.

Pulvermueller, F. & Schoenle, P.W. (1993) Behavioral and neuronal changes during treatment of mixed-transcortical aphasia: a case study. Cognition 48:139-161.

Rockstroh, B., Elbert, T., Canavan, A., Lutzenberger, W., & Birbaumer, N. (1989) Slow cortical potentials and behaviour. Baltimore: Urban & Schwarzenberg.

Roesler, F., Heil, M. & Glowalla, U. (1993) Monitoring retrieval from long-term memory by slow event-related potentials. Psychophysiology 30:170-182.

Saron, C.D. & Davidson, R.J. (1989) Visual evoked potential measures of interhemispheric transfer time in humans. Behavioral Neuroscience 103:1115-1138.

Schuster, H.G. & Wagner, P. (1990) A model for neuronal oscillations in the visual cortex: 1. Mean-field theory and derivation of the phase equations. Biological Cybernetics 64:77-82.

Sheer, D.E. (1984) Focused arousal, 40-Hz EEG, and dysfunction. In Self-regulation of the brain and behavior. T. Elbert, B. Rockstroh, W. Lutzenberger & N. Birbaumer, eds. Berlin: Springer Verlag, pp. 64-84.

Singer, W. (1994) Putative functions of temporal correlations in neocortical processing. In Large scale neuronal theories of the brain. C. Koch & J. Davis, eds. Boston, MA: MIT Press.

Speckmann, E.-J. & Elger, C.E. (1982) Neurophysiological basis of the EEG and of DC potentials. In Electroencephalography - basic principles, clinical applications and related fields. E. Niedermeyer & F. Lopes da Silva, eds. Baltimore, Munich: Urban & Schwarzenberg, pp. 1-13.

Spydel, J.D., Ford, M.R. & Sheer, D.E. (1979) Task dependent cerebral lateralization of the 40 Hz EEG rhythm. Psychophysiology 16:347-350.

Wilson, H.R. & Cowan, J.D. (1973) A mathematical theory of the functional dynamics of cortical and thalamic nervous tissue. Kybernetik 13:35-80.

Volume: 5 (next, prev) Issue: 48 (next, prev) Article: 1 (next prev first) Alternate versions: ASCII Summary