Rhythmic context influences the auditory evoked potentials of musicians and nonmusicians

Abstract

In this study, we investigated how rhythms are processed in the brain by measuring both behaviourally obtained ratings and auditory evoked potentials (AEPs) from the EEG.

We presented probe beats on seven positions within a test bar. Two bars of either a duple- or triple meter rhythm preceded probe beats. We hypothesised that sequential processing would lead to meter effects at the 1/3 and 1/2 bar positions, whereas hierarchical processing would lead to context effects on the 1/3, 1/2 and 2/3 bar positions.

We found that metric contexts affected behavioural ratings. This effect was more pronounced for rhythmic experts. In addition, both the AEP P3a and P3b component could be identified. Though metric context affected the P3a amplitudes, group effects were less clear.

We found that the AEP P3a component is sensitive to violation of temporal expectancies. In addition, behavioural data and P3a correlation coefficients (CCs) suggest that temporal patterns are processed sequentially in nonmusicians but are processed in a hierarchical way in rhythmic experts.

Introduction

The general aim of this study was to investigate how the brain processes rhythmical information and how a mental representation of a rhythm leads to expectancies about events in the near future. This was investigated by means of evoked potentials and by means of behaviourally obtained ratings.

There are several theories concerning the processing of (musical) rhythms. An early paper concerned with rhythm processing is one by Martin (1972). In this paper, Martin proposes a distinction between sequential versus hierarchical coding of temporal sequences. Earlier psychological models concerned with temporal relationships represented relationships between elements as being sequential only. Martin (1972) however proposed that any real time sequence of events might possess a hierarchical internal structure. For example, on a first, sequential level, an underlying invariant, e.g. the beat, is determined. This beat induction is fundamental to the processing of temporal information. One level up, equal subdivisions between beats in either two (as with duple-meters) or in three (as with triple meters) often occur. On a higher level, these subdivisions themselves might be divided in two or three equal parts. In line, the pattern-based, or beat-based model of Povel (1981) also proposes that the first step in the perception of temporal sequences is beat-induction. On a higher level, periods within beats should be either empty; filled with events at equal intervals or filled with events at unequal intervals, provided the subdivision relate as 1:2. The statistical approach of expectancy implies involuntary but relatively long lasting expectancies that are acquired probabilistically: “one expects what is most probable” (Palmer and Krumhansl, 1990). By examining the rhythmic structure of existing musical pieces, expectancy profiles belonging to either a duple- or triple-meter can be estimated. More recent theories like the dynamic attending and the distributed expectancy approach have proposed a dynamic view of rhythm-induced expectancies. The entrained dynamic attending approach thus models attention as an internal oscillatory periodicity. This oscillator can thus predict the maximum expectancy, or when attention in the future is maximal. Oscillators are activated and/or increased by each incoming event. On a lower-order sequential level, the beat will activate such an oscillator. In addition, on a higher-order hierarchical level both beats and subdivisions between beats will activate multiple, possible coupled, oscillators (Large and Kolen, 1992, Large and Palmer, 2002). The distributed expectancy approach (Desain, 1992) models expectancy also in a dynamic way. This theory however bases the mental organization of temporal patterns on time intervals instead of the events. Given a rhythmical sequence, complex expectancy profiles are generated with a maximum on the dominant time-interval and additional expectancies on multiples or subdivisions of this interval.

Though all of the above-mentioned theories differ dramatically in how they model rhythm-induced expectancies, they predict more or less similar expectancies given a certain short rhythm. Also, none of the above-mentioned theories proposes a difference in rhythm-induced expectancies between musicians or naı̈ve listeners. We hypothesize that differences between musicians and nonmusicians will exist such that with an increase in musical experience, there will be an increase in hierarchical levels in which rhythmical sequences can be perceived. Thus, we propose that musical experience will lead to higher-order hierarchical coding of rhythmical patterns instead of a first-order sequential processing of temporal patterns. Since different order levels of rhythm processing lead to different expectancies, in this study we determined if rhythm-induced expectancies were different within rhythmically trained participants compared to musically untrained participants (MUPs).

Evoked potentials (EPs) are small voltage fluctuations resulting from sensory, cognitive, or motor evoked neural activity. These electrical changes are commonly obtained by averaging EEG epochs time-locked to repeated events. EPs consist of components that are typically divided, based on their latency, into exogenous and endogenous components (Coenen, 1995, Naatanen, 1990). It is assumed that early components (<100 ms after stimulus onset) are primarily determined by the physical characteristics of the external stimulus (Blackwood and Muir, 1990), hence labelled the exogenous components. Cognitive aspects of information processing are well known to modulate the later occurring endogenous components (>100 ms after stimulus onset) (Blackwood and Muir, 1990, Gaillard, 1988). For example, it has long been known that expectancy modulates the endogenous EPs (for review, see Polich and Kok, 1995). When expectancy is violated, EPs typically show a large positive wave around 300 ms after onset of the unexpected stimulus, the so-called P300 or P3 (Castro and Diaz, 2001, Johnson and Donchin, 1980, Squires et al., 1977). In line, when expectancy is confirmed, the EP P3 amplitude will be smaller.

We constructed hypotheses about when a following event is maximally expected given a rhythmical sequence, based on either a first-order sequential or a higher-order hierarchical level. Though the above-mentioned theories do not propose that there might be a difference in rhythmical induced expectancies between musicians or naı̈ve listeners, other researchers using EP measurements have reported that musicians show more refined detection of, e.g. pitch (Russeler et al., 2001) and impure chords (Koelsch et al., 2002). In line, we hypothesize that differences between musicians and nonmusicians will exist such that with an increase in musical experience, there will be an increase in hierarchical levels in which rhythmical sequences can be perceived. Since different levels of rhythm processing lead to different expectancies, EPs might be used to determine if expectancies are different within rhythmically trained participants compared to musically untrained participants. Behaviourally obtained ratings only provide a measure of the end product of the processes involved. Recording EPs however allows one to follow the time-course of the different processes involved in musical expectancy (Regnault et al., 2001). Additionally, EP measurements provide a very direct measure not sensitive to response-bias due to, for example group differences in skill (Gaillard, 1988).

In this study, we elicited EPs by presenting probe beats on either the 1/6, 1/4, 1/3, 1/2, 2/3, 3/4 or 5/6 position within a test-bar. Probe beats were preceded by two bars of either a duple- (2/4) or triple (3/4) meter context and a silent bar. Thus, seven different probe beats were presented in two different metric conditions (duple- and triple meter). A diagram of the presented stimulus material is depicted in Fig. 1. The silent bar was introduced to avoid cumulative effects of inhibition on AEP component amplitudes that occur when two or more stimuli are presented with relatively short inter-stimulus intervals (Cardenas et al., 1997, Fitzgerald and Picton, 1981). Thus, for each probe beat, the last three events (lasting more then a second) were identical a in both conditions. This design let to the following hypotheses. Graphical representations of the H₀ and H₁ are depicted in Fig. 2a and b.

H0

The temporal pattern is processed according to a sequential, i.e. first-order interpretation of the pattern. The probe beat in the test bar either continues the preceding, isochronous pattern (e.g. occurs after the same interval as the interval between beats and sub-beats in the preceding pattern) or is a discontinuation of the preceding isochronous pattern (i.e. occurs after a deviant interval as in the preceding pattern). Thus, within the duple-meter trials, only a high expectancy towards probe beats on the 1/2 bar position will arise. In the triple meter trials, a high expectancy towards probe beats on the 1/3 position in the test bar will arise. For these hypotheses, the amount of expectancy is depicted as a Gaussian curve for the duple meter context. For the triple meter context, the curve is skewed, since no probe beats can occur earlier than the beginning of the test bar, but can still occur at the end of the test bar. Thus, according to this hypothesis, a maximal effect of context will appear on the probe beats on the 1/3 position of the test bar (see Fig. 2a). Thus, the estimated, relative values of expectancies towards probe beats (according to their position of occurrence in the test bar) in the duple-meter context can be described as: [1/2]>[1/3,2/3]>[1/4,3/4]>[1/6,5/6], but in the triple meter context as: [1/6]<[1/4]<[1/3]>[1/2]>[2/3]>[3/4]>[5/6].

H1

The temporal pattern is processed in accordance with an extra hierarchical level, taking into account the induced meter, i.e. beats are not only expected to continue the previously presented isochronous pattern, but are also expected to occur on subdivisions and multiples of the interval between beats and sub-beats. In this study, we have modelled expectancy according to this hypothesis as the sum of several Gaussian curves, one around each relevant maximum (Desain, 1992). Thus, within the duple meter trials not only a high expectancy arises on the 1/2 bar position, but also expectancies with respect to the 1/4 and 3/4 positions will arise. Similarly, in the triple-meter trials, expectancies on the 1/3 position, but also on the 2/3 position, and to a lesser extend on the 1/6, 1/2 and 5/6 positions will arise. The estimated relative values of expectancies towards probe beats in the duple-meter context can then be described as: [1/2]>[1/4,3/4]>[1/6,1/3,2/3,5/6], but in the triple meter context as: [1/3,2/3]>[1/6,1/2,5/6]>[1/4,3/4]. According to this hypothesis, a maximal effect of context will appear on the probe beats on the 1/2 position of the test bar (see Fig. 2b).