Predictive Coding of Music-Brain Responses to Rhythmic Incongruity

cortex 45 (2009) 80–92 available at www.sciencedirect.com journal homepage: www.elsevier.com/locate/cortex Special issue: Research report Predictive coding of music – Brain responses to rhythmic incongruity Peter Vuusta,b,*, Leif Ostergaarda, Karen Johanne Pallesena,c, Christopher Baileya,c,d and Andreas Roepstorff a,e a Centre of Functionally Integrative Neuroscience, University of Aarhus, Denmark Royal Academy of Music, Aarhus, Denmark c BioMag Laboratory, Helsinki Brain Research Centre, Helsinki University Central Hospital, Finland d PET Center, Aarhus University Hospital, Denmark e Institute of Anthropology, Linguistics and Archaeology, University of Aarhus, Denmark b article info abstract Article history: During the last decades, models of music processing in the brain have mainly discussed the Received 11 January 2007 specificity of brain modules involved in processing different musical components. We argue Reviewed 29 May 2007 that predictive coding offers an explanatory framework for functional integration in musical Revised 20 July 2007 processing. Further, we provide empirical evidence for such a network in the analysis of Accepted 7 May 2008 event-related MEG-components to rhythmic incongruence in the context of strong metric Published online 14 November 2008 anticipation. This is seen in a mismatch negativity (MMNm) and a subsequent P3am component, which have the properties of an error term and a subsequent evaluation in Keywords: a predictive coding framework. There were both quantitative and qualitative differences in Music the evoked responses in expert jazz musicians compared with rhythmically unskilled non- MEG musicians. We propose that these differences trace a functional adaptation and/or a genetic Meter pre-disposition in experts which allows for a more precise rhythmic prediction. Predictive coding ª 2008 Elsevier Srl. All rights reserved. MMN 1. Introduction Models of music processing in the brain have primarily discussed specificity of brain modules involved in processing musical components. In contrast to language processing, primarily located in the left hemisphere, music processing was originally suggested to be right-lateralized (Luria et al., 1965; Signoret et al., 1987). A more detailed modular viewpoint was recently expressed by Peretz and Coltheart (2003) who demonstrated that anatomically distinct sub-modules were not necessarily confined to one hemisphere to process different aspects of music. The Peretz–Coltheart model, based mainly on lesion studies and on studies of acquired and congenital amusia, emphasizes modular specificity at the expense of brain integration. It adequately accounts for certain aspects of neural musical processing particularly processing of pitch (Liegeois-Chauvel et al., 1998; Mendez, 2001; Peretz et al., 1994). However, it fails, Peretz et al. acknowledge, to fully account for processing of rhythm and meter. This appears problematic, as rhythm and meter are constitutive elements of musical structure, and influence how music is perceived and * Corresponding author. Centre for Functionally Integrative Neuroscience, Aarhus University Hospital, Building 30, Norrebrogade, 8000 Aarhus C, Denmark. E-mail address: pv@pet.auh.dk (P. Vuust). 0010-9452/$ – see front matter ª 2008 Elsevier Srl. All rights reserved. doi:10.1016/j.cortex.2008.05.014 cortex 45 (2009) 80–92 understood (Benjamin, 1984; Dalla and Peretz, 2005; Schmuckler and Boltz, 1994). Recently, Friston (2002) provided a promising model of brain function, in which predictive coding, as a central principle of brain organization, provides a link between segregation and integration (for similar viewpoints see Shepard, 2001; Tononi and Edelman, 1998). The model proposes that the interaction between segregation and integration may be described by predictive coding, interpreted in a hierarchical brain organization whereby lower level brain areas estimate predictions of their expected input based on contextual information through backward connections from higher level areas. A comparison between prediction and actual input produces an error term that, if sufficiently large, will be fed forward to call for an update of the model. This generates a recursive process, which aims at minimizing the difference between input and prediction. As representational capacity of any neuronal assembly in this model is dynamic and context sensitive, it addresses the issue of top– down processing (Frith and Dolan, 1997; Roepstorff and Frith, 2004). The predictive coding model entails that the brain constantly tries to extract structural regularities from the surroundings. This concept is well-established in psychology and neurobiology (Mehta, 2001; Schultz and Dickinson, 2000), and has been successfully applied in several fields, e.g., motor control and social interaction (Blakemore et al., 1998; Wolpert et al., 2003), object perception (Kersten et al., 2004) and visual integration (Rao and Ballard, 1999). In this study, we have employed magnetoencephalography (MEG) to test two hypotheses: (1) that neuronal markers of rhythmic incongruities behave in accordance with a predictive coding framework, (2) that musical competence affects the composition of the neuronal networks involved in the processing of rhythm by affecting the neuronal integration. The human auditory system appears to segregate the auditory environment into meaningful streams according to specific rules, and this forms the basis of a prediction of the near auditory future (Bregman, 1990). As a special case, the rhythmic regularity in music is generated by expectations created in different layers of the musical structure (Bharucha and Stoeckig, 1986; Meyer, 1956; Sloboda, 1985). This depends critically on the timing structure provided by the meter, which is based on a fundamental opposition between strong and weak beats (see, e.g., Cooper and Meyer, 1960; Vuust, 2000). Meter provides the listener with a temporal, hierarchical expectancy structure, underlying the perception of music, in which each musical time-point encompasses a conjoint prediction of timing and strength (Large and Kolen, 1994). When metric expectancy structure is violated, it may elicit strong perceptual responses including sensation of tension (Vuust et al., 2006), shift of attention (Jones and Boltz, 1989) and laughter (Huron, 2004). Violations of meter, especially in music favouring a regular beat, therefore appear particularly well suited as substrate for critical examination of the predictive coding model of brain function. If the predictive coding theory is correct, we hypothesize that meter violation generates an error term at the neural level, the size of which depends on degree of violation. If the violation is sufficiently large, it may cause a subsequent evaluation that involves higher level neuronal structures. The first error term should occur locally, 81 while the putative subsequent evaluation would involve integration across hierarchies of neuronal processing. We therefore created rhythm sequences of increasing rhythmic incongruence and measured brain responses with MEG to test the hypothesis that pre-attentive neural responses to increasing rhythmical incongruity could be identified, and would be congruent with an error term and subsequent evaluation. We hypothesized that rhythmic incongruities would elicit the magnetic counterpart of the mismatch negativity (MMNm), an event-related field (ERF), peaking around 100–200 ms from change onset, an index of pre-attentive detection of change in some repetitive aspect of auditory stimulation (Naatanen, 1992), accompanied by a later component the P3a: usually associated with the evaluation of that change for subsequent behavioral action and believed to indicate activity in a network which contains frontal, temporal and parietal sources (Friedman et al., 2001). According to Winkler et al. (1996), MMN reflects a modification of the pre-attentive model of the acoustic environment. This is caused by the incorporation of a new auditory event that mismatches the actual inferences of the model (the model adjustment hypothesis). This is highly compatible with the predictive coding theory which implies that the error term to unexpected events depends on an interaction between the objective differences in stimulus structure and the degree of detail in the expectancy structure. Musicians are known to have longer and more precise temporal integration windows compared to non-musicians (Russeler et al., 2001), more fine-grained representation of temporal structure (Jongsma et al., 2004) and higher sensitivity when detecting small time changes embedded within simple rhythmic patterns (Jones and Yee, 1997). If the predictive coding theory is correct, then the more detailed expectancy structure in musicians should influence both neuronal markers of the prediction error and the neuronal markers of evaluation. We therefore compared rhythmically unskilled non-musicians with expert jazz musicians. Jazz musicians use challenging rhythmic material in their musical performance and are therefore ideal candidates for identifying putative competence dependent differences in the processing of metric violations. We have previously described a leftward lateralization in musicians compared to nonmusicians when exposed to rhythmically challenging material (Vuust et al., 2005). We here extend the analysis to the P3a component and demonstrate how the findings may be explained by a predictive coding framework. 2. Materials and methods 2.1. Subjects, stimuli and task Nine expert jazz musicians (8 men and 1 woman; mean age ¼ 27.22, SE ¼ 1.68; from the Sibelius Academy of Music, Helsinki, Finland), scoring more than 14 in a modified version of the rhythm imitation test employed at the entry examination for Danish music conservatories, and eight rhythmically unskilled non-musicians (6 men and 2 women; mean age ¼ 24.5, SE ¼ 0.87), scoring less than 3 in the rhythm test, 82 cortex 45 (2009) 80–92 participated in this study, approved by The Ethical Committee of Helsinki University Central Hospital. The rhythm imitation test consists of 30 rhythm sequences falling into three categories presented in a semi-randomized manner: (A) quarter notes and eighth notes, the first note on the down-beat of bar 1 (three sequences), (B) syncopated versions of the above or triplets/16th notes (10 sequences), (C) both 16th notes and triplets or metric displacement (17 sequences). The subjects listened to drum beat sequences (Fig. 1): sIda simple 4-beat rock rhythm; sIIdan alteration of sI introducing a syncopation (metric displacement) which breaks the metric expectancy by replacing a weak beat with a strong beat, without interfering with the music pulse; sIIIdan alteration of sI introducing a beat, incongruent with the underlying temporal grid or meter, hence a stronger violation of musical expectancy. sII may be described as a metrical and musically acceptable syncopation breaking the metric expectancy by replacing a weak beat with a strong beat, challenging the sense of meter without interfering with the music pulse; a well-known stylistic feature in jazz (Kernfeld, 2002). In contrast, the break in sIII from SOAs of 312.5 ms to an SOA of 105 ms did not coincide with any normal subdivision of the beat (triplets or 16th notes) and could therefore be described as a non-metrical and musically unacceptable interruption/ violation easily detectable for all subjects. Using MEG, we recorded brain responses to 600 occurrences of the rhythm sequences pseudo-randomized as follows [frequency of occurrence (f.o.) in parentheses]: sI (30%), sII (30%), sIII (30%). Subjects responded by button presses to the occurrence of either sIu or sId: variations of sI in which the last snare drum beat of the sequence had been tuned up (sIu, f.o.: 5%) or down (sId, f.o.: 5%). This directed their attention to the last part of the rhythm sequences, while the rhythmic deviations occurred in the middle part. Prior to recording, subjects practiced the task and performed a handedness test (Oldfield, 1971). Stimuli were delivered with Presentation (Neurobehavioral Systems, Inc.) through plastic tubes and earpieces. Latency from sound delivery to earplugs (18  2 ms) was subtracted from recorded ERF-latency. Button press responses during recordings were observed to ensure that subjects were performing the discrimination task adequately. After recordings subjects filled in a written questionnaire asking them to rate the stimuli according to how disturbing the sequences appeared, how familiar they seemed and how likely they were to appear in contemporary music. After this, they were verbally debriefed (recorded on tape). 2.2. Data acquisition Neuromagnetic signals were recorded in a magnetically shielded room, using a 306-channel VectorView whole-head MEG system (Elekta-Neuromag, Finland). The 700 ms epochs were set according to stimulus type: A, 4900–5600 ms; B, 4687.5–5387.5 ms; and C, 4380–5080 ms, so that the deviant beat occurred always 100 ms after epoch start, defining a ‘‘time-zero’’ after which all stimuli were physically identical for two bars. A single trial was discarded from the average if any number of the following rejection criteria were met: (i) rectified EOG signal exceeded 100 mV; (ii) rectified MEG of any gradiometer exceeded 3000 fT/cm or (iii) slope of any gradiometer exceeded 6000 fT(/cm)/s. Averaged responses were Rhythm Sequences sI 10 sII 10 sIII 9.7925 s sI HH sII BD/HH 312.5 ms HH SD/HH sIII BD/HH 312.5 ms HH HH: Hi-hat SD: Snare Drum BD: Bass Drum SD/HH BD/HH HH 105 ms Fig. 1 – Stimuli. Top, sequences sI, sII, sIII, prepared in Cubase SX (Steinberg), using realistic broadband sounds: bass drum (BD), snare drum (SD) and hi hat (HH) from LM-7. Arrows indicate time of recording. Recording window was set to 100 ms before and 600 ms after the arrow. Bottom, wave-forms and exact timing of the sounds for each of the sequences. SI is a simple four-beat rock rhythm (functions as a standard in the present design). SII is an alteration of sI that introduces a metric displacement (a subtle rhythmic deviation). SIII is an alteration of sI that introduces a non-rhythmic metric violation (a strong rhythmic deviation). 83 cortex 45 (2009) 80–92 used in subsequent analysis filtered with a pass band of 2–30 Hz using a standard zero-phase FIR filter. Two bipolar EOG channels were used to measure blinking and saccadic eye movements. Electrodes were placed above the left eyebrow, below the lower left eyelid (blinking), and on either brow (saccades). Channels were referenced to the tip of the nose and grounded on the right cheek. Subject’s head position relative to the sensors was measured before and after each session by four head position coils. The head coordinate system was defined by three landmarks on the subject’s head (left and right preauricular points and nasion) whose relative positions were determined using an Isotrak 3D digitizer (Polhemus, Inc., Germany). To relate the dipole fitting locations to macroanatomical brain structures, magnetic resonance images (1 mm3 voxels) were obtained from one subject from each of the two groups, using a 1.5 T Siemens Sonata scanner. 2.3. Data analysis Responses were band pass filtered (2–30 Hz), and a baseline of 50 to 0 ms before the deviation (sII and sIII)/standard (sI) was used. ERFs were analyzed using a subset of 31 pairs of orthogonal planar gradiometers over each hemisphere. Magnetometer channels were discarded due to their poor location specificity and lower signal-to-noise ratio. We defined the mean gradient amplitude (MGA) as follows: sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi o   Xn 2 2 ðdBz=dxÞ þðdBz=dyÞ MGA ¼ 1=N i i where {.}i refers to the ith gradiometer pair in a selection of N. MGA is a measure of the instantaneous amplitude of the tangential gradients of the magnetic field over a selection of N gradiometer pairs. The MGA can (1) be used as a marker of salient input in the brain regions covered by the gradiometer subset used, and (2) reasonably be compared across subjects over a sufficiently large channel selection. For each subject in each hemisphere, brain responses to sI, sII and sIII were measured by comparing the maxima of the MGA (MGAmax) in the intervals (100–170 ms) and (170–250 ms). For both intervals, attempts to estimate the parameters of an equivalent current dipole (ECD) were made separately for each subject, for each hemisphere and condition, at the latency of MGAmax using a spherical head model (Neuromag software xfit). The software gives measures for location, orientation and amplitude of each dipole, as well as test values related to the confidence limits attributable to the estimated parameters. A dipolar pattern, goodness of fit (g.o.f.) > 60%, confidence volume < 500 mm3, and P-value < 0.01, were used as criteria for accepting a dipole. The large number of channels (62 in each hemisphere) was chosen to yield a robust source estimate whilst allowing individual differences in brain shape and head position in the scanner. Bias was minimized by use of the same channels for all subjects. Dipole amplitudes (A) were used to calculate an asymmetry index as (Aleft  Aright)/(Aleft þ Aright). Furthermore, asymmetry indices for both the early (100–170 ms) and the late peak (170–250 ms) were calculated on basis of the MGAmaxs as (MGAmax(left)  MGAmax(right))/(MGAmax(left) þ MGAmax(right)). Locations of MMNms and P3ams were compared (threeway ANOVA) to determine whether different or identical neuronal populations were likely to generate the two responses. Statistical tests were applied to examine group differences with respect to lateralization, latency and amplitude of the MMNm and the P3am. 3. Results 3.1. Behavioral data Expert musicians scored higher in the rhythm imitation test than rhythmically unskilled non-musicians (P < 0.001, Z ¼ 3.54, Table 1). The subjects rated sI–sIII as increasingly disturbing [sI < sII: P ¼ 0.08 (non-significant), Z ¼ 1.74; sII < sIII: P ¼ 0.001, Z ¼ 3.3], decreasingly familiar (sI > sII: P ¼ 0.03, Z ¼ 2.18; sII > sIII: P ¼ 0.001, Z ¼ 3.4) and decreasingly likely to appear in contemporary music (sI > sII: P ¼ 0.004, Z ¼ 2.18; sII > sIII: P ¼ 0.001, Z ¼ 3.4), indicating that the subjects experienced the increasing incongruity of the sequences. Experts and rhythmically unskilled non-musicians did not differ in their rating of sI–sIII with respect to whether sequences were disturbing or likely to appear in contemporary music. Experts however reported to be more familiar with sI and sII but not sIII than the rhythmically unskilled nonmusicians (sI: P ¼ 0.02, Z ¼ 2.27; sII: P ¼ 0.001, Z ¼ 3.18; sIII: P ¼ 0.13, Z ¼ 1.5). This probably reflects that jazz musicians are highly familiar with syncopation, and confirms the fact that sIII is not musically acceptable. Table 1 – Summary statistics of the data collected in the rhythm imitation test and the questionnaire. Behavioral results Scale Rhythm imitation test All 1–30 Experts Unskilled 19.6 (1.1) 1.3 (0.3) Questionnaire ratings Task difficulty 1–5 2.0 (0.2) 1.6 (0.2) 2.5 (0.3) Disturbingness of stimuli (2 to 2) SI 1.4 (0.2) 1.4 (0.7) 1.3 (0.4) SII 0.8 (0.3) 1.0 (0.4) 0.5 (0.5) SIII 0.6 (0.3) 0.7 (0.4) 0.6 (0.5) Familiarity with stimuli SI SII SIII 1–5 Likelihood of appearance 1–5 in contemporary music: SI SII SIII Guess of total number of Stimuli (correct answer ¼ 5) 4.5 (0.2) 3.9 (0.3) 2.6 (0.3) 4.9 (0.1) 4.6 (0.2) 3.1 (0.4) 4.0 (0.3) 3.0 (0.4) 2.1 (0.5) 4.8 (0.1) 4.0 (0.2) 2.3 (0.3) 5.0 (0.0) 4.0 (0.3) 2.1 (0.4) 4.6 (0.5) 4.0 (0.3) 2.5 (1.3) 5.0 (0.3) 5.3 (0.5) 4.6 (0.4) Numbers in parentheses denote standard errors of mean. 84 cortex 45 (2009) 80–92 Interestingly, when asked to guess the total number of stimuli, experts did not perform significantly better than rhythmically unskilled non-musicians. This may indicate that the experts focused just as much on the task (discriminating sIu or sId from sI) as did rhythmically unskilled non-musicians. The results in the following section are presented in the following order: Waveform analyses, MGA-analyses, source localization and analyses of the latencies of the ERFcomponents. (Fig. 5, lower right panel). The response to the congruent beat in sI appeared to be slightly right-lateralized across subjects (P ¼ 0.14) and did not differ between groups (P ¼ 0.73). No difference between groups in lateralization of the P3am was found. In fact, there was a significant right lateralization of the P3am across all subjects and sequences (P ¼ 0.01, Z ¼ 2.6). 3.3. Dipoles 3.2. 3.3.1. MMNm MEG data ERFs to sII and sIII (Figs. 2, 3) showed peaks bilaterally in the intervals 100–170 and 170–250 ms. In keeping with earlier studies we interpret these components as an MMNm (Ford and Hillyard, 1981; Naatanen, 1992) and a subsequent P3am (Escera et al., 2000; Escera et al., 1998; Jongsma et al., 2004). Increased rhythmic incongruence caused greater neuronal responses (Figs. 3 and 4, Table 2), as confirmed by a four-way ANOVA on the MGAs; factors: group of subjects (experts/ rhythmically unskilled), sequence (sI/sII/sIII), hemisphere (left/right), component (MMNm/P3am) (F ¼ 165, P < 0.000001), regardless of hemisphere (interaction between sequence and hemisphere: F ¼ 0.67, P ¼ 0.51). The magnitude of the ERP responses was larger for experts to rhythmic incongruity than for rhythmically unskilled nonmusicians (F ¼ 54.88, P < 0.000001), and there was a significant interaction between group and sequence (F ¼ 7.11, P ¼ 0.001), mainly driven by larger neuronal response in experts to sII (Figs. 2, 3 and 4). Experts showed predominantly left hemispheric responses to sIII and sII as opposed to the rhythmically unskilled non-musicians’ more right-lateralized response (Fig. 3), as indicated by significant interaction between group and hemisphere (F ¼ 5.84, P < 0.05) and the asymmetry index calculated for the MMNms (Fig. 5, Table 2) to the rhythmic incongruence in sII and sIII [F ¼ 18.75, P ¼ 0.001, two-way ANOVA, factors: group of subjects and type of incongruence (sII/sIII)]. This difference in lateralization was more pronounced for the subtle incongruence in sII than for sIII For sIII, a single dipole source was estimated from the data using a restricted inverse solution in all subjects. P3am dipoles were found in all experts in both hemispheres but only for three rhythmically unskilled non-musicians in each hemisphere (Table 3). In the interval 100–150 ms dipolar sources (g.o.f 86%, SE ¼ 2, volume ¼ 46 mm2, SE ¼ 8) resided in the temporal cortex, specifically in the transverse temporal gyrus, near the primary auditory cortex in the two subjects that underwent anatomical imaging (Fig. 6). Dipole amplitudes of the MMN to sIII confirmed stronger neural response in experts than rhythmically unskilled subjects (F ¼ 8.17, P < 0.01, two-way ANOVA on the MGAs, factors: group of subjects and hemisphere), no overall effect of hemisphere (F ¼ 1.06, P ¼ 0.31) but interaction between group and hemisphere (F ¼ 8.46, P < 0.01). Dipole amplitudes in the left hemisphere were significantly larger in experts than in rhythmically unskilled nonmusicians (Table 3, P < 0.01) whereas no significant dipole amplitude difference between groups was observed in the right hemisphere. The asymmetry index confirmed that neural response to sIII was predominantly left-lateralized in experts, but right-lateralized in rhythmically unskilled non-musicians (Fig. 5) (experts: P < 0.05; rhythmically unskilled non-musicians: P < 0.05; difference between groups: P < 0.002). Left hemisphere dipole amplitudes in experts significantly differed from the right hemisphere amplitude in rhythmically unskilled non-musicians (P < 0.05) suggesting increased sensitivity to incongruities as distinguished from merely a change in hemispheric dominance. Expert Unskilled Left Right MMN(m) ft/cm 100 Left ft/cm ft/cm channel 242 expert 100 P3a(m) -100 Right channel 2642 expert 100 ft/cm channel 1613 inept channel 1333 inept 100 sI sII sIII -100 -100 -100 sIII sII sI ms 100 500 Fig. 2 – Evoked responses. Typical time courses of averaged magnetic evoked responses from one rhythmically unskilled non-musician and one expert musician. Signals are recorded at the auditory cortex and plotted separately for sI, sII and sIII. Inspecting the ERFs to sI we found no response (N1), probably due to refraction of the neurons. As a consequence of this, and also because of the strong and early ERFs produced by sII and sIII, difference waves between these and sI were in most cases visually similar to the curves produced by sIII and sII. 85 cortex 45 (2009) 80–92 Right Left sIII: MGA a ft/cm MMN(m) sII: MGA 40 200 400 0 600 ft/cm 200 400 600 ft/cm Experts Unskilled 40 Experts Unskilled 40 10 10 200 400 600 0 ft/cm sI: MGA Experts Unskilled 10 10 0 c ft/cm Experts Unskilled 0 b P3a(m) 40 200 400 600 ft/cm 40 Experts Unskilled Experts Unskilled 40 10 10 0 200 400 0 600 200 400 600 Control Study (sIII) ft/cm ft/cm Experts Unskilled 40 Experts Unskilled 40 MGA d 10 10 0 200 400 0 600 200 400 600 Fig. 3 – Grand means. Grand mean of the MGAs for experts and rhythmically unskilled non-musicians plotted for (a) sIII, (b) sII, and (c) sI. (d) Control study for physical summation in sIII (six experts, five rhythmically unskilled non-musicians). Mean MGA 40.0 For sII, no dipolar sources could be obtained for rhythmically unskilled non-musicians. Left hemispheric MMNmdipole sources could be obtained in six out of nine expert subjects but only in three out of nine experts in the right hemisphere. For experts there was no significant difference in the left hemisphere between localized sources of MMNms to sIII and sII in paired t-tests on each coordinate (<1 mm for each coordinate on average, SE < 4, P > 0.8), and the average distance between MMNms to sIII and sII was less that 8 mm, suggesting similar neuronal sources for MMNms to sII and sIII. f T/cm2 20.0 3.3.2. Experts Unskilled 5.0 sI sII sIII Sequence Fig. 4 – Mean MGAs of experts and unskilled subjects. Mean of the MGAs across hemispheres and peaks, for the two groups of subjects, for each sequence. Error bars denote standard mean of error. P3am P3am dipolar sources were localized in all experts but only half of the rhythmically unskilled non-musicians. In experts the difference between the localization of MMNm and P3am sources was not significant in paired t-tests on each coordinate in each hemisphere. However the average distance was 2.0 cm (SE ¼ 0.4) in the left hemisphere and 2.6 cm (SE ¼ 0.6) in the right. This reflects a large inter-subject variability in the estimate of source localization of P3a. In most subjects it was located near primary auditory cortex, but in some subjects it appeared to have frontal localization while in others the localization estimate was in temporo-parietal cortex. Due to signal strength, it was not possible to make a meaningful 86 cortex 45 (2009) 80–92 Table 2 – Mean gradient amplitude (MGAs) for experts and rhythmically unskilled non-musicians. Latency [ms] MGA [ft/cm] tMGA [ft/cm] Asymmetry index Left Right Left Right sI 100–170 ms Experts Unskilled All 148.8 (6.7) 140.4 (7.2) 144.6 (4.8) 131.0 (7.8) 136.4 (7.7) 133.6 (5.4) 10.1 (1.0) 7.5 (0.6) 8.9 (0.7) 11.6 (1.0) 8.5 (1.1) 10.1 (0.8) 21.8 (1.8) 15.9 (1.4) 19.0 (1.3) 0.068 (0.041) 0.042 (0.066) 0.056 (0.036) 170–250 ms Experts Unskilled All 206.8 (5.9) 200.1 (9.2) 203.6 (5.3) 201.0 (6.9) 210.5 (8.3) 205.5 (5.3) 8.6 (0.9) 7.2 (0.5) 7.9 (0.5) 12.2 (1.6) 7.5 (0.5) 10.0 (1.1) 20.8 (2.2) 14.7 (0.7) 17.9 (1.4) 0.152 (0.058) 0.021 (0.044) 0.090 (0.040) sII 100–170 ms Experts Unskilled All 130.4 (7.0) 140.8 (6.1) 135.3 (4.7) 139.8 (6.1) 140.8 (9.1) 140.3 (5.2) 22.0 (4.5) 6.9 (0.6) 14.9 (3.0) 14.8 (1.6) 11.6 (1.4) 13.3 (1.1) 36.7 (5.7) 18.5 (1.4) 28.2 (3.8) 0.139 (0.086) 0.235 (0.071) 0.037 (0.07) 170–250 ms Experts Unskilled All 204.2 (6.0) 202.8 (11.0) 203.5 (5.9) 198.4 (4.8) 202.6 (10.9) 199.8 (5.6) 14.7 (2.3) 8.4 (0.7) 11.8 (1.5) 14.7 (1.5) 10.1 (0.6) 12.5 (1.0) 29.4 (3.7) 18.5 (0.9) 24.3 (2.4) 0.017 (0.046) 0.097 (0.050) 0.055 (0.035) sIII 100–170 ms Experts Unskilled All 113.2 (4.6) 141.4 (5.2) 126.5 (4.9) 116.9 (3.5) 115.4 (3.8) 116.2 (2.5) 46.2 (5.1) 25.4 (2.9) 36.4 (3.9) 39.0 (4.5) 36.4 (2.9) 37.7 (2.7) 85.2 (8.0) 61.8 (4.9) 74.2 (5.5) 0.083 (0.065) 0.193 (0.073) 0.047 (0.058) 170–250 ms Experts Unskilled All 182.7 (3.0) 204.7 (14.8) 193.1 (7.4) 192.2 (5.7) 208.0 (7.7) 199.6 (5.0) 29.2 (2.3) 15.4 (3.2) 22.7 (2.5) 31.8 (3.3) 19.5 (3.1) 26.0 (2.7) 61.0 (4.8) 34.9 (5.5) 48.7 (4.8) 0.036 (0.042) 0.128 (0.106) 0.079 (0.054) Standard error of mean (SE) is denoted in parentheses. tMGA ¼ MGAleft þ MGAright. The asymmetry index is calculated as (MGAright  MGAleft)/ (MGAright þ MGAleft). LEFT ft/cm RIGHT MMN 40 30 20 sI sII sIII 10 RIGHT 20 Experts Unskilled Experts Unskilled 150 sIII 30 0 Experts Unskilled Experts Unskilled RIGHT P3a 40 10 0 LEFT LEFT ft/cm 50 Amplitude Amplitude 50 ms LEFT RIGHT LEFT 140 sIII 130 sII RIGHT 120 Experts Unskilled sI 110 Experts Unskilled 100 -.4 -.2 .0 .2 Asymmetry Index (MMN) .4 Experts Unskilled Experts Unskilled Latency (MMN) -.4 -.2 .0 .2 .4 Asymmetry Index (MGAs: MMN) Fig. 5 – Top, MGAs at the time of MMNm and the P3am, in the left and right hemisphere for experts and rhythmically unskilled non-musicians. Bottom (left to right), asymmetry indices calculated on the basis of dipole amplitudes; MMNm latency in experts and rhythmically unskilled non-musicians; asymmetry indices calculated on the basis of the MGAs. 87 cortex 45 (2009) 80–92 Table 3 – Dipoles. Summary of localization, dipole moment (amplitude), goodness of fit and confidence volume of the neuronal sources for the MMNm and the P3am for experts and rhythmically unskilled non-musicians. Left hemisphere Right hemisphere No. of Dipole moment Goodness of Confidence No. of subjects Dipole Goodness Confidence moment [nAm] of fit [%] volume [mm2] subjects [nAm] fit [%] volume [mm2] sIII MMN Experts Unskilled 9 8 62.7 25.7 90.9 89.1 42.4 62.9 9 8 37.4 37.7 78.5 87.8 37.1 43.1 P3a Experts Unskilled 9 3 36.3 22.2 77.5 80.0 101.7 340.2 9 3 40.1 21.4 81.9 78.4 94.5 164.6 sII MMN Experts Unskilled 6 – 41.3 – 87.6 – 178.1 – 1 – 11.4 – 66.6 – 231.1 – P3a Experts Unskilled 5 – 24.3 – 79.6 – 242.1 – 3 – 18.6 – 86.9 – 188.1 – triple dipole fit to the data. However, the observed pattern is in accordance with P3am reflecting activity in a network centered in the auditory cortex and encompassing also frontal and parietal areas. 3.4. Latencies MMNm latency to rhythmic incongruity (sII and sIII) in experts was shorter than in rhythmically unskilled subjects (F ¼ 5.20, P < 0.05, three-way ANOVA, factors: group of subjects, type of incongruence and hemisphere). There was significant effect of type of incongruence sII (F ¼ 14.99, P < 0.001) indicating earlier responses to stronger rhythmic incongruence supporting our findings of larger MGAs to stronger rhythmic incongruence (see also Tiitinen et al., 1994). Furthermore there was a significant interaction between group and hemisphere (F ¼ 5.47, P < 0.05) (Table 2). No significant effects were found for P3am. MMNm latency to sIII showed significant effect of group (F ¼ 9.54, P < 0.01, two-way ANOVA, factors: group of subjects and hemisphere), hemisphere (F ¼ 6.60, P < 0.05) and an interaction between group and hemisphere (F ¼ 11.72, P < 0.01). For experts, MMNm latency in the left hemisphere was comparable (P > 0.25) to that of the right hemisphere. However, for rhythmically unskilled non-musicians, left hemisphere latency exceeded the right hemisphere (P < 0.05), and also the left hemisphere of experts (P < 0.001) (Fig. 6). Right hemisphere MMNm latency to sIII did not differ between groups (P > 0.5, Table 2). Fig. 6 – Localization. Dipole localization of MMNms to sIII in one expert and one rhythmically unskilled non-musicians (the same subjects as depicted in Fig. 2). Dipole directions are projected onto the individual coronal, axial and sagittal MR-slices. The relative amplitude (left expert: 60 nAm) is represented by size of arrows. 88 3.5. cortex 45 (2009) 80–92 Control experiment One could hypothetically argue that the peak in the ERF at 110–150 ms measured after the bass drum beat in sIII, resembles an N1 (an obligatory negative deflection on the ERP/ERF, with a latency of around 100 ms, in response to the advent of a sound), due to physical summation of the bass drum beat and the immediately preceding snare drum in our experimental design. To counter this, we ran a control condition in which 13 of the participants listened to the bass drum/hi hat alternatively with a preceding snare drum/hi hat occurring 105 ms before. The distance between two bass drum beats in this condition was 420 ms such that a simple 2/4 would be perceived as the basic meter and that the 105 ms distance between the SD/HH and the BD/HH would be perceived as a 16th note. We observed only small or no responses to the bass drum beat preceded by the snare drum beat and no robust dipolar field patterns in any of the participants, indicating that the part played by physical summation was small compared to the effect of rhythmic incongruity (Fig. 3). 4. Discussion Using MEG, we set out to (1) examine whether the components of neuronal markers of perceived rhythmic incongruity would be consistent with a ‘‘predictive coding’’ framework and (2) test whether subject’s competence affected the composition of the neural response. We found event-related responses to strong rhythmic incongruence (sIII) in all subjects, the MMNm peaking at 110–130 ms and the P3am around 80 ms after the MMNm in expert jazz musicians and some of the rhythmically unskilled subjects, as well as responses to subtle rhythmic incongruence (sII) in most of the expert musicians. The MMNms were localized to the auditory cortices, whereas the P3am showed greater variance in localization between individual subjects. MMNms of expert musicians were stronger in the left hemisphere than in the right hemisphere in contrast to P3ams showing a slight non-significant right lateralization. As we shall argue below, we interpret MMNm and P3am as reflecting an error term generated in the auditory cortex and its subsequent evaluation in a broader network including generators in the auditory cortex as well as higher level neuronal sources. This is in keeping with expectations based on the predictive coding theory and suggests that there is congruence between perceptual experience of rhythmic incongruities and the way that these are processed by the brain. Furthermore, we found enhanced processing of rhythmic deviants in expert musicians compared to rhythmically unskilled non-musicians both at the level of the MMNm and the P3am. MMNm from the left hemisphere in experts to the highly rhythmically incongruent metrical violation had much larger amplitude and peaked 30 ms earlier than for rhythmically unskilled non-musicians. In addition, we observed left lateralization of the MMNm in experts compared to non-musicians to both subtle and strong metric violation, consistent with earlier suggestions of music being left-lateralized in musicians (Altenmuller, 2001; Bever and Chiarello, 1974; Ohnishi et al., 2001), but no lateralization of the P3am. Expert musicians were much more advanced rhythmically and metrically than the rhythmically unskilled non-musicians. We therefore propose that the difference in lateralization and strength of the MMNm in neural response between experts and rhythmically unskilled non-musicians reflects a stronger metrical predictive structure in the experts, which again affects the evaluation of the error as reflected by the stronger P3am. This is in keeping with the predictive coding theory, according to which, the size of the error term is dependent on the prediction. 4.1. Perception of rhythm and predictive coding Music presents itself to the brain as an organised, extended time series, and the significance of each event is played out against this larger temporal structure of expectations, anticipations and tensions, provided by the meter. The interplay between local events and global structure is of paramount importance to music perception and it must, arguably, also be reflected in the brain’s processing of music. The anticipatory model, against which the brain compares incoming input, must be generated on the spot, without determined prior knowledge. One solution to this problem is a hierarchical Bayesian framework that estimates causes through a comparison between estimates and incoming events. In such a ‘‘Bayesian inference machine’’ (Friston, 2002) units of the brain are driven to minimize error by some mechanism which implicitly renders posterior estimates of causes given the data. The brain is therefore a system that is changed by differences between the expectation of input and the actual input on all levels in the hierarchy of neuron assemblies of the brain. The MMN signal appears to have the properties of an error signal in a predictive coding framework. MMNs have been found not only to pattern deviations determined by physical parameters such as frequency (Sams et al., 1985), intensity (Naatanen et al., 1987), spatial localization (Paavilainen et al., 1989), and duration (Naatanen et al., 1989), but also to patterns with more abstract properties (Paavilainen et al., 2001; Van Zuijen et al., 2004). We observed MMNs when metric structure was violated in such a way, that the input could not be predicted by the meter established in the previous bars. Previously MMNms have been found to disruptions of a regular sequence of identical sounds (Ford and Hillyard, 1981; Imada et al., 1993; Nordby et al., 1988). Takegata et al. (2001) have shown that MMN increase in size, when a strong regularity, created by two different sounds, is violated compared with a situation where the two different sounds do not establish a common temporal framework. The present study extends these findings to violations of metric anticipation structure created by the more complex pattern of a drum sequence in keeping with the model adjustment hypothesis of Winkler et al. (1996). One of the strong claims in the predictive coding framework is that ‘‘the specialization of any region is determined both by bottom-up driving inputs and by top–down prediction’’ (Friston, 2002, p. 247). This suggests that an error term fed upwards is a sign to higher areas that something in the predictive process went wrong. We should therefore expect that the error signal, generated locally, will produce effects integrating across brain levels. In the following, we shall argue that the P3a may be an index of such integration across brain levels. We observed a P3am, peaking at around 80 ms following the MMN. The P3a is typically observed in passive listening 89 cortex 45 (2009) 80–92 conditions with salient deviants, and is thought to reflect neural activities involving involuntary attentional orienting. This should be more likely to occur in the present paradigm compared to the more common MMN-paradigms in which subjects are watching video or reading a book. Importantly, the task involved the last part of each stimulus and was temporally separated from the time-points in which the rhythmic deviations occurred. The P3a has been linked to expectancy in general and musical expectancy in particular and is sensitive to violations of metric (Jongsma et al., 2004), melodic (Trainor et al., 2002) and harmonic (Janata, 1995) structure. The P3am dipolar sources in the present study showed large intersubject location variability. We could only localize neuronal sources to the P3am evoked by sIII in expert musicians. Under these conditions, the localizations of the MMNm and the P3am did not differ significantly. However, when comparing MMNm and P3am in individual subjects, some had almost identical dipole localization estimates while the P3am dipole in other subjects were more frontally or parietally located. This finding is compatible with the P3am component reflecting activity in a larger network which ties together components from auditory cortex with parietal and frontal brain regions and where the relative strength of the individual sources varies between individual subjects. However, due to signal strength, we were not able to test this hypothesis directly. In contrast to the MMNm, the P3am in the experts was not left-lateralized. It is therefore very unlikely that the P3am reflects only local activity in the auditory cortex. Previously frontal (Daffner et al., 2000; Schroger et al., 2000), auditory (Alho et al., 1998; Opitz et al., 1999) and temporo-parietal (Downar et al., 2000; Knight et al., 1989) sources have been suggested for the P3a (see also Friedman et al., 2001). We therefore suggest that, in contrast with the MMNm, localized in the auditory cortices, the observed P3am reflects activity in a larger network Feedforward Predictive Coding Higher Level Higher Level Higher Level P P P Experts P3a error error MMN P I LEFT Lower level Auditory Cortex P3a P3a MMN { { Primitive Intelligence MMN error RIGHT Auditory Cortex Asymmetry indices P3a P3a MMN MMN Experts SIII -.4 -.2 .0 .2 Dipole amplitude .4 { { SIII SII SIII Expert Inept SII -.4 -.2 .0 .2 .4 MGA Fig. 7 – Predictive coding. Top, the predictive coding model proposes a specific mode of interaction between lower level brain regions (here the auditory cortices) and higher level neocortical structures. Functional integration among brain systems that employ driving (bottom-up) and backward (top–down) connections mediate this adaptive and contextual specialization, where higher level systems provide a prediction of the inputs to lower level regions and lower regions respond to failures to predict with an error term, which is propagated to higher areas. We suggest that the MMN in primary auditory cortex is an instance of such error signal and that the P3a reflects a functional integration that links the auditory cortex with frontal and parietal brain regions. This allows for solving the conflict between input and prediction via changes in the higher level representations, until the mismatch is ‘‘cancelled’’. Bottom, the difference between the lateralized MMN and the bilateral P3am. 90 cortex 45 (2009) 80–92 that ties together components from the auditory cortex with parietal and frontal brain regions. This is what would be expected if an error signal to a metric prediction were to have any effects in a predictive coding framework. We hence suggest that the P3am is an indication of a neural network that acts on the error signal of the MMNm (Fig. 7). In a recent study of harmonic violation, Koelsch et al. (2000) found an early error signal, the frontally located ERAN (peaking at around 180 ms), elicited to harmonic inappropriate chords in the authentic cadence. Trainor et al. (2002) found a frontal P3a peaking at 300–350 ms to violation of melodic expectancy. We propose to extend our hierarchical predictive coding framework also to this finding. The model suggests that at a given level of processing, a mismatch between prediction and actual activation generates an error signal, which leads to an integration with higher areas. In our experiment, the prediction error to the metric violation originated in the auditory cortex. For more complex patterns, according to the predictive coding model, the processing, and hence the error signal, will move to ‘higher’ areas in the chain of auditory processing. Indeed, in the case of harmonic expectancy the error signal appears to originate in Broca’s area and its right hemisphere homologue (Maess et al., 2001). Thereby the ERAN identified by Koelsch et al. and the P3a identified by Trainor just as the MMNm and the P3am found in our study can be seen as prediction errors followed by an integration into a larger network, irrespective of the different types of expectations violated. This suggests that different anticipations set up in music, at least in terms of metric and harmonic structures but probably also at other levels, may interact with the brain in structurally similar ways, by creating interplays of functional segregation and integration. The larger P3am and the larger MMNm on the left in experts suggest that both the competence of the listener and strength of the musical violation determine whether attention is attracted to the stimulus. Interestingly, jazz musicians responded stronger to the syncopation in sII than did nonmusicians, even though they should be more familiar with such syncopations. We suggest that their expertise is effectuated via a more precise predictive coding process, which relies both on a better top–down propagated model (the meter) for the expected stimuli and a more specific processing in the auditory cortex, particularly on the left. At the core of the present study is a central discussion in the theory and philosophy of music: the disagreement about whether meter is caused by phenomenal accents (Meyer, 1956) in the musical pieces (inputs) or by mental structures (Benjamin, 1984; Palmer and Krumhansl, 1990), not necessarily contained in the music as such. The differences between brain responses of experts and rhythmically unskilled non-musicians strongly indicate that neural operations for meter processing is affected by music competence and/or culture (Drake and Ben El, 2003). Music is thus best seen as a biocultural phenomenon (Cross, 2003) and the concept of the meter is only meaningful in the interaction between music and subject, between the work as such and the mental representation of the underlying musical structure. These mental structures are, however, far from fully understood. Acknowledgements The MEG measurements were conducted at BioMag Laboratory, Helsinki Brain Research Centre, Helsinki University Central Hospital, Finland, and the study was funded by the The Danish National Research Foundation. references Alho K, Winkler I, Escera C, Huotilainen M, Virtanen J, Jaaskelainen IP, et al. Processing of novel sounds and frequency changes in the human auditory cortex: magnetoencephalographic recordings. Psychophysiology, 35: 211–224, 1998. Altenmuller EO. How many music centers are in the brain? Annals of the New York Academy of Sciences, 930: 273–280, 2001. Benjamin WE. A theory of musical meter. Music Perception, 1: 355– 413, 1984. Bever TG and Chiarello RJ. Cerebral dominance in musicians and nonmusicians. Science, 185: 537–539, 1974. Bharucha JJ and Stoeckig K. Reaction time and musical expectancy: priming of chords. Journal of Experimental Psychology, Human Perception and Performance, 12: 403–410, 1986. Blakemore SJ, Goodbody SJ, and Wolpert DM. Predicting the consequences of our own actions: the role of sensorimotor context estimation. Journal of Neuroscience, 18: 7511–7518, 1998. Bregman AS. Auditory Scene Analysis: the Perceptual Organization of Sound. Cambridge, Massachusetts: The MIT Press, 1990. Cooper GW and Meyer LB. The Rhythmic Structure of Music. Chicago: The University of Chicago Press, 1960. Cross I. Music as a biocultural phenomenon. Annals of the New York Academy of Sciences, 999: 106–111, 2003. Daffner KR, Mesulam MM, Holcomb PJ, Calvo V, Acar D, Chabrerie A, et al. Disruption of attention to novel events after frontal lobe injury in humans. Journal of Neurology, Neurosurgery and Psychiatry, 68: 18–24, 2000. Dalla BS and Peretz I. Differentiation of classical music requires little learning but rhythm. Cognition, 96: B65–B78, 2005. Downar J, Crawley AP, Mikulis DJ, and Davis KD. A multimodal cortical network for the detection of changes in the sensory environment. Nature Neuroscience, 3: 277–283, 2000. Drake C and Ben El HJ. Synchronizing with music: intercultural differences. Annals of the New York Academy of Sciences, 999: 429–437, 2003. Escera C, Alho K, Schroger E, and Winkler I. Involuntary attention and distractibility as evaluated with event-related brain potentials. Audiology and Neuro-otology, 5: 151–166, 2000. Escera C, Alho K, Winkler I, and Naatanen R. Neural mechanisms of involuntary attention to acoustic novelty and change. Journal of Cognitive Neuroscience, 10: 590–604, 1998. Ford JM and Hillyard SA. Event-related potentials (ERPs) to interruptions of a steady rhythm. Psychophysiology, 18: 322– 330, 1981. Friedman D, Cycowicz YM, and Gaeta H. The novelty P3: an eventrelated brain potential (ERP) sign of the brain’s evaluation of novelty. Neuroscience and Biobehavioral Reviews, 25: 355–373, 2001. Friston K. Beyond phrenology: what can neuroimaging tell us about distributed circuitry? Annual Review of Neuroscience, 25: 221–250, 2002. Frith C and Dolan RJ. Brain mechanisms associated with top–down processes in perception. Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences, 352: 1221–1230, 1997. Huron D. Music-engendered laughter: an analysis of humor devices in PDQ bach. In Proceedings of the Eighth International Conference of Music Perception and Cognition, Evanston, IL, 2004. cortex 45 (2009) 80–92 Imada T, Fukuda K, Kawakatsu M, Masjiko T, Odaka K, Hayashi M, et al. Mismatch fields evoked by a rhythm passage. Advances in biomagnetism, In The Ninth International Conference on Biomagnetism, Vienna, Austria; 1993. p. 118–9. Janata P. Erp measures assay the degree of expectancy violation of harmonic contexts in music. Journal of Cognitive Neuroscience, 7: 153–164, 1995. Jones MR and Boltz M. Dynamic attending and responses to time. Psychological Review, 96: 459–491, 1989. Jones MR and Yee W. Sensitivity to time change: the role of context and skill. Journal of Experimental Psychology, Human Perception and Performance, 23: 693–709, 1997. Jongsma ML, Desain P, and Honing H. Rhythmic context influences the auditory evoked potentials of musicians and nonmusicians. Biological Psychology, 66: 129–152, 2004. Kernfeld B. The New Grove Dictionary. New York: St. Martin’s Press, 2002. Kersten D, Mamassian P, and Yuille A. Object perception as Bayesian inference. Annual Review of Psychology, 55: 271–304, 2004. Knight RT, Scabini D, Woods DL, and Clayworth CC. Contributions of temporal–parietal junction to the human auditory P3. Brain Research, 502: 109–116, 1989. Koelsch S, Gunter T, Friederici AD, and Schroger E. Brain indices of music processing: ‘‘nonmusicians’’ are musical. Journal of Cognitive Neuroscience, 12: 520–541, 2000. Large EW and Kolen JF. Resonance and the perception of musical meter. Connection Science, 6: 177–208, 1994. Liegeois-Chauvel C, Peretz I, Babai M, Laguitton V, and Chauvel P. Contribution of different cortical areas in the temporal lobes to music processing. Brain, 121: 1853–1867, 1998. Luria AR, Tsvetkova LS, and Futer DS. Aphasia in a composer (V.G. Shebalin). Journal of the Neurological Sciences, 2: 288–292, 1965. Maess B, Koelsch S, Gunter TC, and Friederici AD. Musical syntax is processed in Broca’s area: an MEG study. Nature Neuroscience, 4: 540–545, 2001. Mehta MR. Neuronal dynamics of predictive coding. Neuroscientist, 7: 490–495, 2001. Mendez MF. Generalized auditory agnosia with spared music recognition in a left-hander. Analysis of a case with a right temporal stroke. Cortex, 37: 139–150, 2001. Meyer L. Emotion and Meaning in Music. Chicago: University of Chicago Press, 1956. Naatanen R. Attention and Brain Function. London: Lawrence Erlbaum Ass, 1992: 102–211. Naatanen R, Paavilainen P, Alho K, Reinikainen K, and Sams M. The mismatch negativity to intensity changes in an auditory stimulus sequence. Electroencephalography and Clinical Neurophysiology, 40: 125–131, 1987. Naatanen R, Paavilainen P, and Reinikainen K. Do event-related potentials to infrequent decrements in duration of auditory stimuli demonstrate a memory trace in man? Neuroscience Letters, 107: 347–352, 1989. Nordby H, Roth WT, and Pfefferbaum A. Event-related potentials to time-deviant and pitch-deviant tones. Psychophysiology, 25: 249–261, 1988. Ohnishi T, Matsuda H, Asada T, Aruga M, Hirakata M, Nishikawa M, et al. Functional anatomy of musical perception in musicians. Cerebral Cortex, 11: 754–760, 2001. Oldfield RC. The assessment and analysis of handedness: the Edinburgh inventory. Neuropsychologia, 9: 97–113, 1971. Opitz B, Mecklinger A, Friederici AD, and Von Cramon DY. The functional neuroanatomy of novelty processing: integrating ERP and fMRI results. Cerebral Cortex, 9: 379–391, 1999. Paavilainen P, Karlsson ML, Reinikainen K, and Naatanen R. Mismatch negativity to change in spatial location of an auditory stimulus. Electroencephalography and Clinical Neurophysiology, 73: 129–141, 1989. 91 Paavilainen P, Simola J, Jaramillo M, Naatanen R, and Winkler I. Preattentive extraction of abstract feature conjunctions from auditory stimulation as reflected by the mismatch negativity (MMN). Psychophysiology, 38: 359–365, 2001. Palmer C and Krumhansl CL. Mental representations for musical meter. Journal of Experimental Psychology, Human Perception and Performance, 16: 728–741, 1990. Peretz I and Coltheart M. Modularity of music processing. Nature Neuroscience, 6: 688–691, 2003. Peretz I, Kolinsky R, Tramo M, Labrecque R, Hublet C, Demeurisse G, et al. Functional dissociations following bilateral lesions of auditory cortex. Brain, 117: 1283–1301, 1994. Rao RP and Ballard DH. Predictive coding in the visual cortex: a functional interpretation of some extra-classical receptivefield effects. Nature Neuroscience, 2: 79–87, 1999. Roepstorff A and Frith C. What’s at the top in the top–down control of action? Script-sharing and ‘top–top’ control of action in cognitive experiments. Psychological Research, 68: 189– 198, 2004. Russeler J, Altenmuller E, Nager W, Kohlmetz C, and Munte TF. Event-related brain potentials to sound omissions differ in musicians and non-musicians. Neuroscience Letters, 308: 33–36, 2001. Sams M, Paavilainen P, Alho K, and Naatanen R. Auditory frequency discrimination and event-related potentials. Electroencephalography and Clinical Neurophysiology, 62: 437–448, 1985. Schmuckler MA and Boltz MG. Harmonic and rhythmic influences on musical expectancy. Perception and Psychophysics, 56: 313–325, 1994. Schroger E, Giard MH, and Wolff C. Auditory distraction: eventrelated potential and behavioral indices. Clinical Neurophysiology, 111: 1450–1460, 2000. Schultz W and Dickinson A. Neuronal coding of prediction errors. Annual Review of Neuroscience, 23: 473–500, 2000. Shepard RN. Perceptual-cognitive universals as reflections of the world. Behavioral and Brain Sciences, 24: 581–601, 2001. Signoret JL, Van Eeckhout P, Poncet M, and Castaigne P. Aphasia without amusia in a blind organist. Verbal alexia-agraphia without musical alexia-agraphia in braille. Revue Neurologique (Paris), 143: 172–181, 1987. Sloboda J. The Musical Mind. Oxford: Oxford University Press, 1985. Takegata R, Syssoeva O, Winkler I, Paavilainen P, and Naatanen R. Common neural mechanism for processing onset-to-onset intervals and silent gaps in sound sequences. Neuroreport, 12: 1783–1787, 2001. Tiitinen H, May P, Reinikainen K, and Naatanen R. Attentive novelty detection in humans is governed by pre-attentive sensory memory. Nature, 372: 90–92, 1994. Tononi G and Edelman GM. Consciousness and the integration of information in the brain. Advances in Neurology, 77: 245–279, 1998. Trainor LJ, McDonald KL, and Alain C. Automatic and controlled processing of melodic contour and interval information measured by electrical brain activity. Journal of Cognitive Neuroscience, 14: 430–442, 2002. Van Zuijen TL, Sussman E, Winkler I, Naatanen R, and Tervaniemi M. Grouping of sequential sounds – an eventrelated potential study comparing musicians and nonmusicians. Journal of Cognitive Neuroscience, 16: 331–338, 2004. Vuust P. Polyrhythm and Metre in Modern Jazz – a Study of the Miles Davis’ Quintet of the 1960’ies (Danish). Aarhus, Denmark: Royal Academy of Music, 2000. Vuust P, Pallesen KJ, Bailey C, Van Zuijen TL, Gjedde A, Roepstorff A, et al. To musicians, the message is in the meter pre-attentive neuronal responses to incongruent rhythm are left-lateralized in musicians. Neuroimage, 24: 560–564, 2005. 92 cortex 45 (2009) 80–92 Vuust P, Roepstorff A, Wallentin M, Mouridsen K, and Ostergaard L. It don’t mean a thing. Keeping the rhythm during polyrhythmic tension, activates language areas (BA47). Neuroimage, 31: 832–841, 2006. Winkler I, Karmos G, and Naatanen R. Adaptive modeling of the unattended acoustic environment reflected in the mismatch negativity event-related potential. Brain Research, 742: 239–252, 1996. Wolpert DM, Doya K, and Kawato M. A unifying computational framework for motor control and social interaction. Philosophical Transactions of the Royal Society of London. Series B, Biological Sciences, 358: 593–602, 2003.