Cognitive Brain Research 12 (2001) 131–144 www.elsevier.com / locate / bres Research report Simultaneously active pre-attentive representations of local and global rules for sound sequences in the human brain ´ ´ a , *, Istvan ´ Czigler a , Elyse Sussman b , Istvan ´ Winkler a,c Janos Horvath a Institute of Psychology, Hungarian Academy of Sciences, H-1394 Budapest, P.O. Box 389 Szondi u. 83 /85, Budapest, Hungary b Department of Otolaryngology, Albert Einstein College of Medicine, New York, NY, USA c Cognitive Brain Research Unit, Department of Psychology, University of Helsinki, Helsinki, Finland Accepted 27 February 2001 Abstract Regular sequences of sounds (i.e., non-random) can usually be described by several, equally valid rules. Rules allowing extrapolation from one sound to the next are termed local rules, those that define relations between temporally non-adjacent sounds are termed global rules. The aim of the present study was to determine whether both local and global rules can be simultaneously extracted from a sound sequence even when attention is directed away from the auditory stimuli. The pre-attentive representation of a sequence of two alternating tones (differing only in frequency) was investigated using the mismatch negativity (MMN) auditory event-related potential. Both localand global-rule violations of tone alternation elicited the MMN component while subjects ignored the auditory stimuli. This finding suggests that (a) pre-attentive auditory processes can extract both local and global rules from sound sequences, and (b) that several regularity representations of a sound sequence are simultaneously maintained during the pre-attentive phase of auditory stimulus processing. 2001 Elsevier Science B.V. All rights reserved. Theme: Neural basis of behavior Topic: Learning and memory: systems and functions Keywords: Auditory event-related potentials; Mismatch negativity; Regularity representation; Temporal organization 1. Introduction In everyday life, sounds are usually encountered within sequences of inter-related auditory events. Therefore, the sequential organization of sensory information is a key element of auditory perception. Organization is achieved through the formation of ‘links between parts of the sensory data’ (Bregman, 1990, p. 47 [4]), which gives rise to the segmentation of the auditory environment. The processes associating sounds according to various auditory organization principles could take place at different stages of information processing. Bregman’s [4] theory of auditory stream segregation assumes that several of the possible ways of linking sounds together are examined in parallel during the early, possibly pre-attentive phases of auditory information processing. We tested the hypothesis that both *Corresponding author. ´ E-mail address: horvath@cogpsyphy.hu (J. Horvath). the local and global types of links are formed during the pre-attentive stages of auditory processing. Regular sequences of sounds (i.e., non-random) can usually be characterized by more than one rule. One may classify these rules on the basis of their temporal scope. Local rules describe relations between temporally adjacent stimuli, global rules describe relations between temporally non-adjacent stimuli [4]. Simple sound sequences can be described in terms of both local and global rules. For example, the local rule for a sequence of two alternating tones (‘ . . . ABABAB . . . ’) would say that A is always followed by B and B is always followed by A. One possible global rule describing the same sequence would maintain that every second tone is A while every other is B. Another categorization of rules can be made on the basis of whether a rule is concrete (i.e., enables extrapolating from a specific sound to another specific sound) or abstract (i.e., enables extrapolating from one class of sounds to 0926-6410 / 01 / $ – see front matter 2001 Elsevier Science B.V. All rights reserved. PII: S0926-6410( 01 )00038-6 132 ´ et al. / Cognitive Brain Research 12 (2001) 131 – 144 J. Horvath another class of sounds; abstract rules define relationships between sets of sounds). In the above example of the two alternating tones, both rules are concrete as they apply to the two specific tones of the alternating sequence. An abstract rule describing (frequency) alternation, in general, would state that the two neighbors of a sound in the sequence have either both higher or both lower pitch than the sound itself. Finally, a distinction which is relevant for theories of auditory information processing: a rule may apply to stimuli as integrated events (‘event-rules’) or only to certain sound features (‘feature-rules’). Local and global, concrete and abstract rules may apply either to integrated sound events or to levels of some sound feature(s). Our first pair of examples were event-rules; the abstract rule example was a feature rule. A global concrete feature rule, which also applies to simple alternation, is that the time from the onset of a tone to the onset of the next identical tone is constant at a given stimulus onset asynchrony (SOA) value. The systematic formation of associations between sounds in regular sequences can be regarded as a neural representation of a rule. The question of what kind of rules are represented for a given sound sequence in the absence of focused attention by the human auditory system can be studied using an event-related brain potential (ERP) component termed the mismatch negativity (MMN) [14] (for recent reviews, see Refs. [28,16]). MMN is elicited by sounds that violate some regularity of the preceding auditory stimulus sequence, whether or not the subject’s attention is focused on the auditory modality [13]. It has been established that the discriminative MMN-generating process involves an auditory sensory memory representation of the preceding regular sounds [12,21,16]. Winkler et al. [36] suggested that MMN elicitation results from the incoming sound mismatching the one(s) extrapolated from the preceding auditory stimulus sequence. Therefore, if a sound elicits the MMN, one can infer that some regularity in the preceding sound sequence was pre-attentively detected (this regularity being violated by the MMNeliciting auditory stimulus). Thus MMN enables one to determine what rules are pre-attentively represented by the auditory system for a given sound sequence. The regularities which have been shown to be detected during the pre-attentive phase of auditory information processing range from simple repetition rules (e.g., Ref. [14]) to sensory trends [33] and regularities of complex sequential or spectro-temporal sound patterns (e.g., Refs. [18,25]). MMNs elicited by violations of concrete as well as abstract rules (e.g., Refs. [14] and [23], respectively), event as well as feature rules (e.g., Refs. [37] and [8], respectively) have already been observed. However, no previous study has made a direct test of whether global rules can be represented nor did they distinguish between repeating patterns and the global structure of non-adjacent tones. 2. Experiments The aim of the present study was to investigate whether (a) global rules can be pre-attentively extracted and (b) different kinds of rules (such as global and local, concrete and abstract) can be simultaneously represented by the pre-attentive auditory information processing system. The main test sequence (Experiment 1) was a simple alternation of two tones (see Fig. 1). As was described above, such a sequence can be characterized by several different rules, amongst them both local and global ones. Previous studies (e.g., Refs. [1,18]) demonstrated that occasional repetitions of one of the two alternating tones (‘ . . . ABABAA . . . ’) elicit the MMN even when subjects ignore the auditory stimuli. This means that some regularity or regularities describing this tone alternation were pre-attentively represented in the auditory system. However, these previous studies did not ask the question whether the regularities underlying the elicitation of the observed MMNs were local- or global-rule representations: the tone repetition deviation, which was examined, simultaneously violates both the local and global rules. Testing what kind of rules of alternation lie behind the observed MMN responses is made possible by the fact that extrapolations based on local vs. global rules come up with different results for the sound following an occasional sound repetition. According to the local rule, A is followed by B and B is followed by A; therefore, if the local rule returns after a sound repetition, this repetition should be followed by changing to the other sound (‘ . . . ABABAAB . . . ’). In contrast, according to the global rule, each sound is identical to the sound two back in the sequence; therefore, for the global rule to return, the sound following a repetition should be identical to the repeated sound (‘ . . . ABABAAA . . . ’). Since sounds violating Fig. 1. The stimulus sequence presented in Experiment 1. The two alternating tones differed only in frequency (1000 and 1100 Hz). The different types of deviants are marked by different fillings. ´ et al. / Cognitive Brain Research 12 (2001) 131 – 144 J. Horvath some pre-attentively detected regularity elicit the MMN component, the ERP responses to the two possible continuations of the alternating sequence following an occasional stimulus repetition will indicate whether the preattentive MMN-generating process was based on the local or a global alternation rule. If both types of rules are simultaneously used in the MMN-generating process, neither or both continuations should elicit the MMN (depending on whether violation of one rule supersedes compliance with another rule or not). 2.1. Experiment 1 2.1.1. Subjects and procedure Thirteen paid subjects (nine females; 16–24 years of age) with normal hearing participated in the experiment. Subjects gave informed consent after the nature of the experiment was explained to them. The subject was comfortably seated in an acoustically shielded room. He / she was instructed to ignore the sounds and read a selfselected book. The experiment consisted of ten stimulus blocks, each 8.4 min long, with 1–2 min rest between consecutive blocks. 2.1.2. Stimuli Stimuli were pure sinusoidal tones presented binaurally through headphones with an SOA of 500 ms at 70 dB (SPL). The stimulus duration was 100 ms (including 5 ms rise and fall times). Two tones differing only in frequency (1000 and 1100 Hz) were alternated (‘ . . . ABAB . . . ’). Stimulus blocks consisted of 1010 tones. In each block, the pattern of alternation was violated 40 times by repeating either one of the tones with equal probability. Henceforth, tone repetitions (which violate both local and global rules) will be referred to as ‘both-rules deviants’. After half of the both-rules deviants, the sequence continued with a change between the two tones: ‘ . . . ABABAAB . . . ’ or ‘ . . . BABABBA . . . ’. This type of continuation violates the global, but not the local rule (see above); hence we term the change following a single tone repetition ‘globalrule deviant’ (see Fig. 1). Following the other half of the both-rules deviants, the sequence continued with another repetition of the same tone: ‘ . . . ABABAAA . . . ’ or ‘ . . . BABABBB . . . ’. This type of continuation violates the local, but not the global rule (see above); hence we term the second consecutive tone repetition ‘local-rule deviant’. Each both-rules deviant was preceded by at least 5 regularly alternating tones. Although this puts some restriction on the randomness of the sequence, the low probability of deviants allow sufficient number of variations that the length of possible cycles in the stimulus sequence would still fall, by far, outside the capacity of auditory memory. Overall, responses to 103405400 bothrules deviants and half as many local- and global-rule deviants were recorded from each subject. 133 2.1.3. EEG recording EEG was recorded (DC-40 Hz, sampling rate 250 Hz, Synamp amplifiers, NeuroScan EEG recording system) with Ag /AgCl electrodes placed on five locations on the midline (Fpz, Fz, Cz, Pz, Oz), by the two mastoids (Lm, Rm), and at the one- and two-third points of the arc connecting Fz and the mastoids over both hemispheres (L1, L2 and R1, R2 on the left and right scalp, respectively). The common reference electrode was attached to the tip of the nose. The horizontal electro-oculogram was recorded using a bipolar configuration between electrodes positioned near the outer canthi of the two eyes. The vertical EOG was recorded with a bipolar configuration between electrodes placed above and below the right eye. 2.1.4. Data analysis EEG was off-line bandpass filtered in the 1.5–20 Hz range and epoched between 290 and 490 ms from stimulus onset. The first ten epochs of each block and epochs containing an amplitude change exceeding 75 mV on any EEG or EOG recording were rejected from further analysis. ERP responses were averaged separately for ‘standards’ (regularly alternating tones following at least two cycles of undisturbed alternation), both-rules deviants, local-rule deviants, global-rule deviants, and the first change following a local-rule deviant (‘ABABAAAB’ or ‘BABABBBA’) which we term ‘repetition-rule deviant’ (see Fig. 1) as it may elicit an MMN if the preceding three tones established a repetition rule (see Refs. [5,27,36]). Responses elicited by the two different tones (A and B) were collapsed in each of the above cases. To estimate the full MMN amplitude (see Ref. [13]) for statistical analysis, ERPs were re-referenced to the left mastoid (Lm). Figures will, however, show the nosereferenced responses and differences separately for Fz and Lm, for assessing the expected polarity inversion of the MMN response. The amplitude measurements were referred to the average amplitude in the 90 ms pre-stimulus interval. The MMN amplitude was calculated by subtracting the standard response from each of the four deviant responses. Mean MMN amplitudes were measured from 24 ms wide intervals centered at the negative peaks observed in the 100–200 ms post-stimulus interval of the grandaverage deviant-minus-standard differences. As the MMN peak latency varies considerably with experimental parameters like the amount of deviance and the complexity of the stimulation (see, e.g., Refs. [13,32]) and the temporal extent of the component is relatively short, this widely used method provides the best signal to noise ratio for detecting the presence of the component. One-sided dependent t-tests were used to verify the presence of MMN. The MMN amplitudes elicited by the both-rules, localrule, and global-rule deviants were analyzed by one-way repeated measures ANOVA. Additivity of MMN responses elicited by local- and global-rule deviants were tested by one-way repeated measures ANOVA between the response 134 ´ et al. / Cognitive Brain Research 12 (2001) 131 – 144 J. Horvath to both-rules deviants and the sum of responses to localand global-rule deviants (in the MMN range of the bothrules deviant response). The scalp topographies of the MMNs elicited by the local- and global-rule deviants were compared using two-way repeated measures ANOVA on normalized data (to eliminate main effects that could obscure possible interaction effects) in accordance with the recommendations of McCarthy and Woods [10]; see also Ref. [20]). 2.1.5. Results Fig. 2 (left column) summarizes the grand-average nose- referenced frontal (Fz) ERP responses to the standard and the four deviant tones. A frontally negative waveform peaking in the 100–200 ms post-stimulus latency range can be observed for each of the deviant-minus-standard difference curves (Fig. 2, right column). A reversal in polarity of this negative deflection can be seen at the Lm electrode site. The inversion of polarity below the Sylvian fissure is a characteristic feature of the MMN component [13], whereas the N2b component, which can also appear in this latency range shows no such polarity inversion. Grand-averaged MMN peak latency and amplitude measurements at Fz (re-referenced to Lm) and t-test results for Fig. 2. Frontal (Fz) grand-average ERP responses (nose-referenced) to standard and deviant tones (left column) and the corresponding deviant-minusstandard difference curves at Fz and Lm (right column) in Experiment 1. Shading indicates the full (Lm-referenced) MMN component. Responses to the 4 different types of deviants are shown in different rows. First row: both-rules deviant (‘ . . . ABABAA . . . ’); 2nd row: local-rule deviant (‘ . . . ABABAAA . . . ’); 3rd row: global-rule deviant (‘ . . . ABABAAB . . . ’); 4th row: repetition deviant (‘ . . . ABABAAAB . . . ’). The y-axis marks the moment when the pattern-ending tone (marked by bold in the examples above) was delivered. For each type of deviant, responses were collapsed across the two analogous patterns (obtained by exchanging the roles of A and B). ´ et al. / Cognitive Brain Research 12 (2001) 131 – 144 J. Horvath 135 Table 1 Grand-average frontal (Fz, re-referenced to Lm) MMN peak latencies and mean amplitudes elicited by the four different deviants in Experiment 1 a Deviant type Stimulus-pattern Peak latency (ms) Peak amplitude (mV) t (df512) Both-rules Local-rule Global-rule Repetition-rule . . . ABABAA . . . . . . ABABAAA . . . . . . ABABAAB . . . . . . ABABAAAB . . . 172.3 156.9 154.5 166.5 20.92 21.68 21.07 21.21 24.94** 25.91** 23.66* 24.59** (3.9) (3.5) (3.5) (5.0) (0.18) (0.29) (0.29) (0.26) a Standard errors of the mean are given in parentheses. The deviant tone for each deviant type is typeset in bold in the stimulus patterns. Standards were regularly alternating tones following at least two cycles of undisturbed alternation. Both deviant and standard responses are collapsed across the two analogous patterns (obtained by exchanging the roles of A and B). * P,0.01. ** P,0.001. all four deviants are given in Table 1. When a change between the two tones was preceded by one or two tone repetitions (Fig. 2, 3rd and 4th rows, respectively), an earlier, frontally negative deviant-minus standard difference wave was also elicited, which was peaking close to 100 ms from stimulus onset (latency5100.067.7 ms [6S.E.M], amplitude520.9360.28, t(12)523.34, P, 0.01 and 103.165.1 ms, 20.7360.18 mV, t(12)53.98, P,0.001, for the global-rule and repetition deviants, respectively). These early negative differences reflect an increase of the N1 amplitude compared with that elicited by regular alternation. Since both the global-rule and repetition deviants are more separated in time from the preceding identical tone (1400 and 1900 ms, offset to onset) than the tones of the regular alternation (900 ms), some of the frequency-specific N1-generator neuronal elements become less refractory and, therefore, respond more vigorously (see further Ref. [15]). The MMN amplitudes across conditions (both-rules, local-rule and global-rule) did not differ (one-way repeated measures ANOVA, F(2,24)52.78, P,0.1). The MMN amplitude elicited by the both-rules deviant was smaller than the sum of the MMN amplitudes elicited by the localand global-rule deviants in the MMN-range of the bothrules deviant (F(1,12)510.51, P,0.01). There was no difference in the scalp distribution of the MMNs elicited in the local- and global-rules conditions (two-way repeated measures ANOVA with factors of Condition [local- vs. global-rule deviant] and Electrode [Fz, Cz, L1, Lm, R1, Rm] on normalized data; F(5,60)50.68, P,0.6) 2.1.6. Discussion The elicitation of MMN by both the local- and globalrule deviants suggests that local and global rules are simultaneously represented in the pre-attentive auditory information processing system. Deviants violating the local but not the global rule could only elicit the MMN, if the local rule was maintained even though subjects were engaged in a task that was unrelated to the auditory stimulus sequence. The fact that no N2b appeared in the deviant responses confirms that subjects did not actively attend the sound sequences, as in this case, deviants should have elicited the N2b component [13,22,25]. If we could conclude that the global-rule deviants elicited MMN because they violated the global but not the local rule, the simultaneous representation of the two types of rules would be proven. However, four alternative explanations can also account for the MMN elicited by global-rule deviants: 1. Repeating a sound twice or more has been shown to build up a representation of a repetition rule (i.e., the recurrence of the same sound; [5,27,36]). Therefore it is possible that even a single repetition of a sound may be already sufficient for a subsequent change to elicit the MMN component. This alternative explanation would be compatible with a present result showing that a change following two repetitions (three presentations) of one of the test tones (the repetition deviant) elicited the MMN (Fig. 2, 4th row). Because the repetition deviant violated neither the global nor the local rule, it is very likely that the MMN elicited by this deviant was based on violation of a repetition rule which was formed as the auditory system was presented with 3 consecutive identical tones. Therefore, the global-rule deviant, which is a change following one repetition (2 presentations) of one of the tones might reflect a similar phenomenon. 2. The MMN elicited by the global-rule deviant might be an analogue of the MMNs observed for standard stimuli which immediately followed a deviant [24,19,36]. In the present case, the first repetition in the alternating sequence is a deviant with respect to the local rule, and the subsequent change (the globalrule deviant) is a standard with respect to this local rule. Although the standard-following-a-deviant MMN phenomenon has only been observed for repetition rules, one cannot reject this alternative on the basis of the Experiment 1 results. 3. It is also possible that the MMNs observed in Experiment 1 were elicited by infrequent changes in the temporal schedule of stimulus presentation. One may regard the alternating sequence as two separate repeti- 136 ´ et al. / Cognitive Brain Research 12 (2001) 131 – 144 J. Horvath Fig. 3. Stimulus sequences presented in Experiment 2. Conditions are shown in different rows. Top row: Condition 1, sequences were equiprobably randomized from twenty tones differing only in frequency. Middle row: Condition 2, increase in frequency was followed by decrease in frequency and vice versa; the size of frequency change was random. Bottom row: Condition 3, two tones with 1000 and 1100 Hz frequencies were presented in an alternating sequence with random (250–750 ms) SOA. tive series of tones (one higher, the other lower) which are interleaved with each other. Repeating a tone violates the regular SOA of the series to which it belongs (e.g., ‘ . . . ABABAA . . . ’ can be regarded as ‘ . . . A A AA . . . ’, an occasional reduction of the regular SOA of the A series). Indeed, had the rate of presentation been fast enough for the A and B tones to form separate streams [4], this is how one would perceive this event. Similarly, all of the present deviants can be regarded as violations of the otherwise isochronously presented higher / lower-tone series (if these are processed separately). MMN was shown to be elicited by occasionally shortening the SOA in an otherwise isochronously presented sound sequence [7,17]. It is, therefore, possible that the temporal properties of the two sub-series are maintained preattentively and violations of the alternation of two isochronously presented tones are detected as deviance from these temporal regularities. 4. Finally 1 , the tones immediately preceding the globalrule deviant form a micro-sequence in which the occurrence of the repeated tone is higher than that of the other tone (e.g. in the case of the preceding 4 tones –‘ABAA’-A is presented in 75% of the stimuli). Micro-sequences with unequal probabilities for the two tones have been shown to elicit MMN in 50–50% 1 We thank an anonymous reviewer of the manuscript for suggesting this alternative explanation. (global probability) random sequences of two tones [24]. Thus this alternative explanation suggests that the MMN elicited by the global-rule deviant could be the result of the unequal probabilities of the two alternating tones in the micro-sequence preceding the global-rule deviant. 2.2. Experiments 2 and 3 Experiments 2 and 3 were designed to test the four alternative explanations. In Condition 1 of Experiment 2 (see Fig. 3, top row), random tones with varying frequencies were presented. If repeating a sound once already establishes a representation of a repetition rule for this sound then the stimulus change immediately following an occasional tone repetition should elicit the MMN. In contrast, if the stimulus change immediately following a repetition does not elicit an MMN, then Alternative 1 can be ruled out. In Condition 2 of Experiment 2 (see Fig. 3, middle) a sequence of tones whose frequencies alternated in an abstract way were used. Neighbors of a tone were either both lower or higher in frequency than the tone itself. The sequence was like this: ‘ . . . Higher–Lower–Higher– Lower–Higher–Lower . . . ’ The amount of frequencychange between consecutive tones was set to be random with successive tones being easily distinguishable from each other. This way, forming a global rule of the alternation (i.e. to infer the frequency relationship between ´ et al. / Cognitive Brain Research 12 (2001) 131 – 144 J. Horvath temporally non-adjacent tones) was impossible, only the abstract local rule could be encoded. This frequency alternation regularity was then occasionally violated by presenting two successive tones whose frequencies differed from their predecessors in the same direction (i.e. ‘Higher– Higher’ or ‘Lower–Lower’). If the abstract frequency alternation rule was pre-attentively detected, such violations should elicit MMN. The tone following these deviants returned to the alternation rule (i.e. ‘Higher–Higher–Lower’ or ‘Lower–Lower–Higher’), therefore, it constituted a ‘standard’ following a local-rule only deviant. If this return to alternation elicits the MMN then the ‘first standard after a deviant’ explanation (Alternative 2) is correct. Since this sequence cannot be described in terms of global rules, the first standard following a deviant situation is not confounded by global rule violation. Therefore, if the ‘standard’ following a ‘deviant’ does not elicit an MMN in this case then Alternative 2 can be ruled out. In Condition 3 of Experiment 2 (see Fig. 3, bottom row), a simple alternation between the two tones was presented using a random SOA. This procedure prevents the alternation-violation deviants from simultaneously violating the SOA-regularity of the assumed separate high and low sub-series because, with the random-SOA presentation, the SOA in these sub-series also becomes variable. If occasional tone repetitions do not elicit an MMN in the random-SOA sequence then one should conclude that the constancy of SOA is an essential part of the representation of simple tone alternation and that the auditory system might have detected the alternation violations in Experiment 1 as violations of the SOA regularity in the high and low sub-series (Alternative 3). In contrast, if MMN is elicited by alternation violations even when the tones are presented with a random SOA then the violation of SOAconstancy is not a necessary prerequisite of the detection of alternation violations and, therefore, Alternative 3 can be ruled out. Finally, in Experiment 3 (see Fig. 4) random tones with varying frequencies were presented (similarly as in Condition 1 of Experiment 2). Micro-sequences with an ‘AXAAB’ or ‘AAXAB’ pattern (A±X, A±B, and X±B) were occasionally presented within the random stimulus sequence. These patterns had three out of the four tones preceding the B (probe) tone identical, but no three 137 identical tones in a row, just like it was before the global-rule deviant in Experiment 1. If the probe tone elicits an MMN in this situation, then Alternative 4 is correct. In contrast, if these ‘3 out of 4 identical’ microsequences result in no MMN elicitation by the probe tone, then Alternative 4 can be ruled out. It should be noted that this situation is somewhat more favorable for MMN elicitation by the probe than that was for the global-rule deviant in Experiment 1 as, unlike the tones in the regular alternation, the probe does not appear frequently within the stimulus sequence (it is simply one of the random frequencies used to compose the sequence). The alternation conditions of Experiment 2 realize different generalizations of the simple tone-alternation regularity. Therefore, alternation deviants in Experiment 2 also test whether these different generalized regularities are detected and represented pre-attentively. 2.2.1. Subjects and procedure Fourteen paid subjects with normal hearing (12 females, 16–22 years of age), none of whom participated in Experiment 1, participated in Experiment 2. Due to extensive artifacts (e.g. alpha-activity) resulting in the rejection of a large number of trials, three subjects’ data were discarded. Subjects were instructed to ignore the sounds and read a self-selected book. Eleven paid subjects with normal hearing (9 females, 18–23 years of age), none of whom participated in Experiment 1 or 2, participated in Experiment 3. Subjects were instructed to ignore the sounds and watch a selfselected subtitled movie without sound. One subject reported counting the stimuli despite the instruction to disregard them; the data of this subject was excluded from the analysis. 2.2.2. Stimuli Stimuli were 100 ms long (including 5 ms rise and fall times) pure sinusoid tones presented binaurally through headphones at 70 dB (SPL). In Experiment 2 stimulus blocks consisted of 1010 tones. In Condition 1 of Experiment 2, sequences were equiprobably randomized from twenty tones differing only in frequency (see Fig. 3, top row). Tone frequencies ranged from 800.0 to 1278.9 Hz with proportionally equal, 2.5% frequency steps. Consecutive tones of the sequences Fig. 4. The stimulus sequence presented in Experiment 3. The sequences were equiprobably randomized from twenty tones differing only in frequency, with infrequent five-tone long micro-sequences occurring randomly. The first four tones of the critical micro-sequences are of two kinds of tones differing in their frequencies. One of the tones is presented three times (marked by gray shading) providing a relatively high local probability for this tone. The fifth tone, called the probe, is marked by black color. 138 ´ et al. / Cognitive Brain Research 12 (2001) 131 – 144 J. Horvath differed by at least two frequency-steps. SOA was 500 ms. The two stimulus blocks of this condition contained 100 stimulus repetitions each. Stimulus repetitions were preceded by at least five stimulus changes. In Condition 2 of Experiment 2, increase in frequency was followed by a decrease in frequency and vice versa. The stimulus sequences were randomized using forty different tones covering, with equal 2.5% steps, the 625– 1678 Hz frequency range (see Fig. 3, middle row). Frequency change between consecutive tones was equiprobably two, three, or four frequency steps. The SOA was 500 ms. In the four blocks of this condition, frequencychange alternation was violated fifty times each, (twentyfive times repeating frequency increase and twenty-five times frequency decrease, the extent of frequency-change being randomly two, three or four steps). Alternation violations were preceded by at least five regular frequencychanges. In Condition 3 of Experiment 2, two tones with 1000 and 1100 Hz frequencies were presented in an alternating sequence (see Fig. 3, bottom row). The SOA was randomized with a uniform distribution in the 250–750 ms range. In each of the four blocks of this condition, alternation was violated fifty times by tone repetition (twenty-five times ‘AA’ and twenty-five times ‘BB’). Following the repetition, the sequence continued with a change. Repetitions were preceded by at least five alternating tones. In Experiment 3, stimulus blocks were constructed similarly to those of Condition 1 of Experiment 2, except that they consisted of 890 tones. Each of the eight stimulus blocks in this experiment contained 44 micro-sequences, in which a tone occurred exactly three times within four consecutive stimuli. Half of these micro-sequences had the pattern of ‘AXAA’, the other half ‘AAXA’, A and X differing by at least two frequency steps. These microsequences were followed by a tone (the probe) that differed both from the A and X tones by at least two frequency steps. The critical micro-sequences were preceded by at least two random frequency changes. 2.2.3. EEG recording and data analysis The parameters of EEG recording and data analysis were identical to that in Experiment 1. In Condition 1 of Experiment 2, responses to tone repetitions and the first change following these repetitions were analyzed and compared with the standard response (i.e., the response to stimulus change with no deviant in the 5 preceding positions of the sequence). In Condition 2 of Experiment 2, the responses to repeated frequency increases and decreases (collapsed across deviants), and the response elicited by the first regular sound change following these deviants (the ‘standard after the deviant’) were compared with the response to regular sound change following at least five regular changes. In Condition three of Experiment 2, the response to tone repetitions (alternation deviant) were compared with the standard response (i.e., the response to regular tone alternation with no tone repetition within the preceding five positions). For control purposes, we also compared the responses elicited by standards separated from the previous identical tone by a short SOA (i.e., the time between the two ‘A’s in ‘ . . . ABA . . . ’ ranging from 500 to 750 ms) with standards separated from the preceding identical tone by a medium-duration SOA (ranging from 875 to 1125 ms). Because the time between two consecutive identical standards was the sum of two random SOAs (both taken from the 250–750 ms uniform distribution), 43.7% of the standards belong to the medium-duration SOA range, and 12.5% of the standards belong to the short SOA range. If deviants (repetition violations) elicit an MMN because of their lower-than-average temporal separation from the preceding identical tone (which was between 250 and 750 ms for deviants, as opposed to the regular 500 to 1500 ms separation between two consecutive identical standard tones) then the short-SOA standards could be expected to elicit a similar MMN response. In Experiment 3 responses to the probe tones (the ones following the critical microsequences) were compared with the standard response (i.e., the response to stimulus change separated by at least two tones from the previous probe stimulus). 2.2.4. Results Figs. 5–8 present the grand-average nose-referenced frontal (Fz) ERP responses obtained in Experiments 2 and 3. Table 2 summarizes the results of Experiments 2 and 3. Infrequent tone repetition embedded in the sequence with continuously changing frequency (Condition 1 of Experiment 2) elicited a frontally negative difference wave (compared with the response to regular stimulus change; Fig. 5, 1st row; see also Table 2). The change following the repeated tone also elicited a frontally negative difference response (Fig. 5, 2nd row) peaking 109.465.6 ms from stimulus onset with an amplitude of 20.6960.29 mV (t(10)522.41, P,0.05). This negative difference was identified as an increased N1 response due to its early peak latency which corresponded to the similar N1 findings of Experiment 1 (in which the increased N1 and the MMN waves could be separately identified in the responses recorded for changes following one or more tone repetitions; see Fig. 2, 3rd and 4th rows). In Condition 2 of Experiment 2, the second consecutive frequency increase or decrease elicited a frontally negative difference response (compared with the ‘standard’ frequency-change alternation response). This response was identified as an abstract frequency-alternation MMN (Fig. 6, 1st row, also see Table 2). No significant ERP difference was obtained between the ‘standard’ tone following frequency-change alternation deviants and the regular ‘standard’ response (Fig. 6, 2nd row). Note that the (statistically not significant) difference that can be seen at about 170 ms on the left mastoid recording is in the ‘wrong’ (negative) direction for an MMN response. ´ et al. / Cognitive Brain Research 12 (2001) 131 – 144 J. Horvath 139 Fig. 5. Frontal (Fz) grand-average ERP responses (nose-referenced) to standard and test tones (left column) and the corresponding test-minus-standard difference curves at Fz and Lm (right column) in Condition 1 (random frequency change) of Experiment 2. Shading indicates the full (Lm-referenced) MMN component. First row: tone repetition; 2nd row: change after repetition (the tone following a repeated tone). In Condition 3 of Experiment 2, tone repetitions (Fig. 7, 1st row) elicited the MMN response (see Table 2). ShortSOA standards (see definition in EEG recording and data analysis) did not elicit a significant differential response compared with the medium-SOA standards (Fig. 7, 2nd row). The slight (statistically not significant) difference between 120 and 220 ms could not be an MMN, as its polarity is opposite to that of the MMN component. Fig. 6. Frontal (Fz) grand-average ERP responses (nose-referenced) to standard and test tones (left column) and the corresponding test-minus-standard difference curves at Fz and Lm (right column) in Condition 2 (frequency-change alternation) of Experiment 2. Shading indicates the full (Lm-referenced) MMN component. First row: alternation deviant (two consecutive frequency increases or decreases); 2nd row: standard following an alternation deviant (a frequency decrease following two consecutive frequency increases or an increase following two decreases). 140 ´ et al. / Cognitive Brain Research 12 (2001) 131 – 144 J. Horvath Fig. 7. Frontal (Fz) grand-average ERP responses (nose-referenced) to standard and test tones (left column) and the corresponding test-minus-standard difference curves at Fz and Lm (right column) in Condition 3 (random-SOA tone alternation) of Experiment 2. Shading indicates the full (Lm-referenced) MMN component. First row: alternation deviant (tone repetition); 2nd row: comparison between alternation standards separated from the preceding identical tone by short (500–750 ms) vs. medium (875–1125 ms) intervals. In Experiment 3, probes following the critical four-tone micro-sequences in which three out four tones were the same, did not elicit any observable difference compared with the ‘regular’ frequency change response (Fig. 8). Note that the frontal (Fz) negative differences identified in the 100–300 ms post-stimulus interval were all accompanied by differential positive waves over the mastoid leads. This suggests that N2b was not elicited by the deviant and probe stimuli in Experiments 2 and 3. 2.2.5. Discussion No MMN was elicited by a stimulus change following an infrequent tone repetition within the sequence of randomly varying tones (Condition 1 of Experiment 2). This suggests that a single repetition is not sufficient for setting up a repetition-regularity representation. Therefore, the first alternative explanation of the MMN elicited by the global-rule deviant in Experiment 1 (see the Discussion of Experiment 1) can be ruled out. This result seems to be at odds with those obtained by Winkler et al. [34], who presented stimulus trains separated by 9.5 s intervals with the tone frequencies changing from train to train. In this study, MMN was obtained to tones presented in the third position of those trains in which the third tone differed in frequency from the common frequency of the first two tones. Although this result indicated the possibility ‘that the mismatch process can be elicited by a deviant stimulus following only two presentations of the standard’, the authors noted that ‘however, carryover of the transient memory trace from the previous stimulus train did not Fig. 8. Frontal (Fz) grand-average ERP responses (nose-referenced) to standard and probe tones (left column) and the corresponding probe-minus-standard difference curves at Fz and Lm (right column) in Experiment 3. Probe tones followed four-tone long micro-sequences in which one tone occurred three times without two consecutive repetitions. ´ et al. / Cognitive Brain Research 12 (2001) 131 – 144 J. Horvath 141 Table 2 Summary of results from Experiments 2 and 3 a Experiment and condition Stimulus-pattern Peak latency (ms) Peak amplitude (mV) t (df510) Exp. 2, Cond.1 . . . ABCC . . . b . . . ABCCD . . . b . . . HLHH . . . c . . . HLHHL . . . c . . . ABAA . . . Short-SOA standard d . . . AXAAB . . . or . . . AAXAB . . . e 151.3 (8.3) – 156.4 (5.4) – 161.1 (4.9) – – 21.08 (0.25) – 20.61 (0.24) – 20.67 (0.30) – – 24.25** – 22.50* – 22.19* – – Exp. 2, Cond.2 Exp. 2, Cond.3 Exp. 3 a When MMN was elicited, the grand-average frontal (Fz, re-referenced to Lm) MMN peak latencies and mean amplitudes (standard errors of the mean in parentheses) are given. In Conditions 1 and 2 of Experiment 2 and in Experiment 3, standards were regular tone changes following at least five regular changes. In Condition 3 of Experiment 2, standards were regularly alternating tones following at least two cycles of undisturbed alternation; only the ones separated by an SOA between 875 and 1125 ms were compared with the Short-SOA-standards. Both deviant and standard responses were collapsed across the analogous patterns. b A±B, B±C, C±D. c H stands for a Higher, L stands for a Lower tone with respect to the tone immediately preceding it. d Two identical tones separated by an SOA between 500 and 750 ms. e A±X, A±B, B±X. * P,0.05. ** P,0.001. allow a definitive test of this issue’ (both sentences are from Winkler et. al 1993, p. 411 [34]). Thus we can conclude that the MMN obtained by Winkler et al. [34] in the 3rd-position of their roving-frequency trains was probably elicited with respect to the regularity set up by the previous train (the carryover effect shown by Winkler et al. [34]), whereas the formation of a new repetitive regularity requires more repetitions of the same stimulus (see below for further discussion). Although tones following a repetition did not elicit an MMN, the infrequent tone repetitions themselves elicited a negative deflection whose characteristics were similar to that of the MMN component. The infrequent tone repetitions appearing in a sequence whose regular feature is stimulus (frequency) change violate this ‘change-regularity’. On this basis, the frontally negative ERP response elicited by infrequent tone repetitions may be regarded as an MMN component. Similar results have been obtained ¨ by Wolff and Schroger [40]. In contrast, Ritter et al. [22] found no MMN to infrequent (20% probability) tone repetitions in a sequence randomized from five different tones. The lack of a repetition-specific ERP response in Ritter et al.’s study [22] might have resulted from the high probability of repetition (20%), the low number of different frequencies used (as it has been shown that the auditory system can probably maintain more than five frequently recurring tones simultaneously; see Ref. [37]), and / or the longer SOA in Ritter et al.’s design. An alternative explanation of the differential ERP response elicited by infrequent tone repetitions embedded in a sequence of frequently changing tones would suggest that this component reflects a process for detecting constancy in an ever-changing environment. It could be argued that there are a number of regularities represented in the auditory system for other acoustic features (e.g. SOA, intensity, and tone duration are constant) but there is none for frequency. The lack of any frequency-regularity representation may be a distinct state of the pre-attentive auditory system that makes it sensitive to regularities emerging in the frequency dimension. In this case, the observed negative deflection can be considered as a correlate of the emergence of a new regularity representation when no regularity representation was previously available. However, two arguments can be put forward against this idea. First, the regularity representation emerges after a single tone repetition, then MMN should have been elicited by a different tone following this repetition. However, no MMN was elicited by the change following this repetition. Second, according to the ‘new regularity explanation’, a similar negative deflection should be elicited by the second of two identical tones presented after a long silent period. However, Cowan et al. [5] found no such response when comparing trains starting with two identical versus two different tones (trains were preceded by 11–15 s of silence and tone frequencies changed from train to train — the ‘roving-standard’ condition). Therefore, it seems more likely that ‘frequencychange’ can be pre-attentively represented as an abstract feature-rule, with respect to which infrequent tone repetitions elicit the MMN. The lack of N2b elicitation by tone repetitions supports the view that these instances were pre-attentively detected as the attentive detection of tone repetitions has been shown to elicit the N2b component [22]. In Condition 2 of Experiment 2 the return to the local alternation rule after an alternation violation did not elicit MMN. No global alternation rule could be formed for this sequence. Therefore, the second alternative explanation of the MMN elicited by the global-rule deviant in Experiment 1 (see the Discussion of Experiment 1) can also be ruled out (i.e. that this MMN is analogous to the MMNs 142 ´ et al. / Cognitive Brain Research 12 (2001) 131 – 144 J. Horvath observed for standard stimuli following a deviant in simple repetitive tonal sequences, see [24,19,36]). Importantly, however, infrequent violations of this local-rule-only alternation (two consecutive frequency increases or decreases) elicited the MMN. This result demonstrated the ability of the auditory system to pre-attentively represent frequency-change-alternation, a generalized form of simple frequency alternation. In Condition 3 of Experiment 2, infrequent tone repetitions elicited the MMN in a simple alternating sequence presented with random SOA, whereas ‘standard’ (regularly alternating) tones presented with a short SOA did not. This result rules out the alternative explanation suggesting that infrequent repetitions and subsequent structural violations in an alternating sequence of two tones elicit MMN because they violate the regular temporal separation between consecutive identical tones (see Discussion of Experiment 1). Results of this condition demonstrate that the MMNs elicited by infrequent structural violations in a sequence of two regularly alternating tones at least partly reflect deviation from a genuine structural representation of these sequences. The present results provide a further important implication: the constancy of SOA is not a necessary pre-requisite for the pre-attentive detection of alternation. This means that also this generalization of alternation can be represented even in the absence of focused attention. However, previous studies demonstrated that the temporal aspects of the stimulation are intimately linked with the perceived structure of auditory stimulus sequences. By changing the timing of stimulus delivery one can modify the perception of sound sequences. For example, speeding up the presentation rate of the present sequence of alternating tones would make the low and high tones form separate streams; see Ref. [4]. A number of recent studies suggest that the temporal and structural features of sound sequences are processed, and probably also encoded, together during pre-attentive auditory processing [2,3,32] as well as that the schedule of stimulus delivery can determine the elicitation of the MMN component by violations of sequential (structural) regularities [29,30,39]. The present results do not contradict the notion of integration between the temporal and structural features of sound sequences. In fact, the lack of MMN in the Short-SOA-standard responses in Condition 3 showed that temporal and structural regularities of a given sequence refer to the same organization of this sequence. (MMN elicitation by Short-SOA alternation standards would have suggested that the segregated [separate low and high sequences] organization of the sequence was maintained simultaneously with the integrated [low–high alternation] organization.) In Experiment 3, the probe tones did not elicit an observable MMN. These probes followed micro-sequences that modeled the ratio between the two tones in the micro-sequence immediately preceding the global-rule deviant in Experiment 1. This result rules out the alternative explanation, which suggests that the MMN elicited by the global-rule deviant resulted from unequal local stimulus probabilities (see Discussion of Experiment 1). 3. General discussion The aim of the present study was to determine (a) whether global rules (i.e., relations between temporally non-adjacent stimuli) can be represented pre-attentively and (b) whether different types of rules describing the same sound sequence are maintained simultaneously in the absence of focused attention. The present results showed that global-rule representations of auditory stimulus sequences could be formed pre-attentively. This suggests that the auditory system is able to associate non-adjacent sounds within the larger structure of a sequence, even when one’s attention is directed away from the sounds. The second main conclusion from the present results is that at least two different rules can be pre-attentively abstracted from a simple alternating tone sequence. Furthermore, incoming sounds were checked against the representation(s) of both of these rules. Previous studies have shown that rules for the frequent return of two (possibly even more) sounds can be maintained at the same time (e.g., Ref. [38]). It is also true that rules describing the repetition of a stimulus event include all featurerepetition rules as well as the repetition of feature conjunctions, as deviations in different auditory features and / or conjunction of features elicit separate MMNs within the same sequence (e.g. Refs. [6,9,26,31]). Feature-repetition rules can also coexist with more complex structural rules of the sequence. Winkler & Czigler [35] presented a sequence of two regularly alternating tones that differed only in frequency. Infrequent tones having the frequency of the tone preceding them and a duration which was shorter than the common duration of the two ‘standard’ tones elicited two successive MMNs (one to the break of frequency alternation and another to duration deviance) demonstrating that both alternation and the constancy of duration were simultaneously represented during pre-attentive auditory processing. Corroborating evidence was obtained for alternation between the two ears and frequency-intensity conjunctions by [32]. From these and the present results it is clear that a large number of rules (feature and event rules, local and global rules) are simultaneously detected and represented during the preattentive analysis of the auditory environment. The present results further suggest that concrete and abstract rules of the same sequence are maintained in parallel. Violations of a regular frequency-change alternation (the abstract variant of frequency alternation) elicited the MMN (see Fig. 6). As the regular alternation of two tones differing only in ´ et al. / Cognitive Brain Research 12 (2001) 131 – 144 J. Horvath frequency is a special case of this general class of alternations, it is possible that the general rule is also established for simple alternations. Furthermore, the present results are compatible with the notion that redundant structural rules describing the same sound sequence are encoded together. In Experiment 1, the sum of the responses to local- and global-rule deviants significantly differed from the response to the both-rules deviant. This indicates that the responses are not additive. In fact, the MMN amplitudes elicited by these three different deviants did not significantly differ from each other. However, it must be noted that the present paradigm was not optimal for testing the additivity of the local- and global-rule MMNs, as the both-rules deviant followed a somewhat different sequence than either the local or the global-rule deviant. Additionally, the scalp distributions of the MMNs were similar, which suggests that these responses are generated by at least partially overlapping neuronal populations. The present conclusions suggesting simultaneous representations of multiple redundant regularities for the same stimulus sequence are fully compatible with Bregman’s [4] assumption of the existence of several pre-attentive processes that analyze the auditory input in parallel. These processes provide the basis for the organization of the acoustic environment. The present results also strongly support the hypothesis that the auditory system maintains a complex neural model of the acoustic environment even in the absence of focused attention [36]. As the auditory environment continuously changes, the regularities extracted from it require constant monitoring, lest they become outdated. Deviant auditory events thus serve a double role: (1) they carry new, potentially important information (i.e., information that could not be extrapolated from the preceding sequence) and, therefore, may require further evaluation; (2) they mark changes in the regularities of the auditory environment (i.e., instances when new rules may supersede some previously detected rules) and, therefore, initiate modifications in the representation of rules. The MMN-generating process is probably involved in both of these functions [11,34,36]. The present results provided a demonstration of the adaptability of this system (cf. [36,16]). Two successive repetitions breaking the alternation (three consecutive presentations of the same tone) established a repetition-rule representation, as was shown in Experiment 1 (see Fig. 2). However, a single repetition embedded in a sequence of tones with randomly changing frequencies was not sufficient for creating a new repetition rule representation (see Fig. 4). Comparable results have been obtained in previous studies [5,27,36]. These results demonstrate that the auditory system quickly adapts to changes in the regularities of the acoustic environment but rejects chance occurrences. A minimal prerequisite of three events for the activation of a new rule representation is probably an optimum between 143 the need for fast adaptation and for efficiency (i.e., creating short-living, possibly false, rule representations would tax the available resources and reduce the utility of the model). Acknowledgements This research was supported by the Hungarian National Research Fund (OTKA T022800) and the National Institutes of Health Grant (R55 DC04263). References [1] C. Alain, D.L. Woods, K.H. Ogawa, Brain indices of automatic pattern processing, NeuroReport 6 (1994) 140–144. [2] C. Alain, A. Achim, D.L. Woods, Separate memory-related processing for auditory frequency and patterns, Psychophysiology 36 (1999) 737–744. [3] C. Alain, F. Cortese, T.W. Picton, Event-related brain activity associated with auditory pattern processing, NeuroReport 10 (1999) 2429–2434. [4] A.S. Bregman, Auditory Scene Analysis, MIT Press, Cambridge MA, 1990. ¨¨ ¨ [5] N. Cowan, I. Winkler, W. Teder, R. Naatanen, Memory prerequisites of mismatch negativity in the auditory event related potentials (ERP), J. Exp. Psychol. Learn. Mem. Cogn. 19 (1993) 909–921. [6] D. Deacon, J.M. Nousak, M. Pilotti, W. Ritter, C.M. Yang, Automatic change detection: douse the auditory system use representations of individual stimulus features or gestalts? Psychophysiology 35 (1998) 413–419. [7] J.M. Ford, S.A. Hillyard, Event-related potentials (ERPs) to interruptions of a steady rhythm, Psychophysiology 18 (1981) 322–330. [8] H. Gomes, R. Bernstein, W. Ritter, H.G. Vaughan Jr., J. Miller, Storage of feature conjunctions in transient auditory memory, Psychophysiology 34 (1997) 712–716. ¨ [9] S. Levanen, R. Hari, L. McEvoy, M. Sams, Responses of the human auditory cortex to changes in one vs. two stimulus features, Exp. Brain Res. 97 (1993) 177–183. [10] G. McCarthy, C.C. Wood, Scalp distributions of event-related potentials: An ambiguity associated with analysis of variance models, Elecroencephalogr. Clin. Neurophysiol. 62 (1985) 203–208. ¨¨ ¨ [11] R. Naatanen, Selective attention and stimulus processing: Reflections in event related potentials, magnetoencephalogram and regional cerebral blood flow, in: M.I. Posner, O.S.M. Marin (Eds.), Attention and Performance XI, Lawrence Erlbaum, Hillsdale, NJ, 1985, pp. 355–373. ¨¨ ¨ [12] R. Naatanen, The role of attention in auditory information processing as revealed by event-related potentials and other brain measures of cognitive functions, Behav. Brain Sci. 13 (1990) 201–288. ¨¨ ¨ [13] R. Naatanen, Attention and Brain Function, Lawrence Erlbaum, Hillsdale NJ, 1992. ¨¨ ¨ ¨ [14] R. Naatanen, A.W. Gaillard, S. Mantysalo, Early selective attention effect on evoked potential reinterpreted, Acta Psychol. 42 (1978) 313–329. ¨¨ ¨ [15] R. Naatanen, T.W. Picton, The N1 wave of human electric and magnetic response to sound: a review and an analysis of the component structure, Psychophysiology 24 (1987) 375–425. ¨¨ ¨ [16] R. Naatanen, I. Winkler, The concept of auditory stimulus representation in cognitive neuroscience, Psychol. Bull. 125 (1999) 826– 859. [17] H. Nordby, W.T. Roth, A. Pfefferbaum, Event-related potentials to 144 [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] ´ et al. / Cognitive Brain Research 12 (2001) 131 – 144 J. Horvath time-deviant and pitch-deviant tones, Psychophysiology 25 (1988) 249–261. H. Nordby, W.T. Roth, A. Pfefferbaum, Event-related potentials to breaks in sequences of alternating pitches or interstimulus intervals, Psychophysiology 25 (1988) 262–268. J.M.K. Nousak, D. Deacon, W. Ritter, H.G. Vaughan Jr., Storage of information in transient auditory memory, Cogn. Brain Res. 4 (1996) 305–317. T.W. Picton, S. Bentin, P. Berg, E. Donchin, S.A. Hillyard, R. Johnson Jr., G.A. Miller, W. Ritter, D.S. Ruchkin, M.D. Rugg, M.J. Taylor, Guidelines for using event-related potentials to study cognition: recording standards and publication criteria, Psychophysiology 37 (2000) 127–152. W. Ritter, D. Deacon, H. Gomes, D.C. Javitt, H.G. Vaughan Jr., The Mismatch negativity of event-related potentials as a probe of transient auditory memory: a review, Ear Hear. 16 (1995) 52–67. W. Ritter, P. Paavilainen, J. Lavikainen, K. Reinikainen, K. Alho, M. ¨¨ ¨ Sams, R. Naatanen, Event-related potentials to repetition and change of auditory stimuli, Elecroencephalogr. Clin. Neurophysiol. 83 (1992) 306–321. ¨ J. Saarinen, P. Paavilainen, E. Schroger, M. Tervaniemi, R. ¨¨ ¨ Naatanen, Representation of abstract attributes of auditory stimuli in the human brain, NeuroReport 3 (1992) 1149–1151. ¨¨ ¨ M. Sams, K. Alho, R. Naatanen, Sequential effects in the ERP in discriminating two stimuli, Biol. Psychol. 17 (1983) 41–58. ¨ E. Schroger, An event-related potential study of sensory representations of unfamiliar tonal patterns, Psychophysiology 31 (1994) 175–181. ¨ E. Schroger, Processing of auditory deviants with changes in one versus two stimulus dimensions, Psychophysiology 32 (1995) 55– 65. ¨ E. Schroger, On the detection of auditory deviants: a preattentive activation model, Psychophysiology 34 (1997) 245–257. ¨ E. Schroger, Measurement and interpretation of the mismatch negativity, Behav. Res. Methods Instrum. Comput. 30 (1998) 131– 145. E. Sussman, W. Ritter, H.G. Vaughan Jr., An investigation of the [30] [31] [32] [33] [34] [35] [36] [37] [38] [39] [40] auditory streaming effect using event related brain potentials, Psychophysiology 36 (1999) 22–34. ¨¨ ¨ E. Sussman, I. Winkler, W. Ritter, K. Alho, R. Naatanen, Temporal integration of auditory stimulus deviance as reflected by the mismatch negativity, Neurosci. Lett. 264 (1999) 161–164. ¨¨ ¨ R. Takegata, P. Paavilainen, R. Naatanen, I. Winkler, Independent processing of changes in auditory single features and feature conjunctions in humans as indexed by the mismatch negativity, Neurosci. Lett. 266 (1999) 109–112. ¨¨ ¨ R. Takegata, P. Paavilainen, R. Naatanen, I. Winkler, Pre-attentive processing of simple and complex acoustic regularities: a mismatch negativity additivity study, Psychophysiology 38 (2001) 92–98. ¨¨ ¨ M. Tervaniemi, S. Maury, R. Naatanen, Neural representations of abstract stimulus features in the human brain as reflected by the mismatch negativity, NeuroReport 5 (1994) 844–846. ´ ¨¨ ¨ I. Winkler, N. Cowan, V. Csepe, I. Czigler, R. Naatanen, Interactions between transient and long-term auditory memory as reflected by the mismatch negativity, J. Cogn. Neurosci. 8:5 (1996) 403–415. I. Winkler, I. Czigler, Mismatch negativity: deviance detection or the maintenance of the ‘standard’, NeuroReport 9 (1998) 3809–3813. ¨¨ ¨ I. Winkler, G. Karmos, R. Naatanen, Adaptive modeling of the unattended acoustic environment reflected in the mismatch negativity event-related potential, Brain Res. 742 (1996) 239–253. I. Winkler, P. Paavilainen, K. Alho, K. Reinikainen, M. Sams, R. ¨¨ ¨ Naatanen, The effect of small variation of the frequent auditory stimulus on the event-related brain potential to infrequent stimulus, Psychophysiology 27 (1990) 228–235. ¨¨ ¨ I. Winkler, P. Paavilainen, R. Naatanen, Can echoic memory store two traces simultaneously? A study of event-related brain potentials, Psychophysiology 29 (1992) 337–349. ¨ I. Winkler, E. Schroger, N. Cowan, The role of large-scale perceptual organization in the mismatch negativity event-related brain potential, J. Cogn. Neurosci. 13 (2001) 59–71. ¨ Ch. Wolff, E. Schroger, Activation of the pre-attentive change detection system by tone repetitions with fast stimulation rate, Cogn. Brian Res. 10 (2001) 323–327.