Psychological Principles of Grouping

A phenomenon already noticed in early listening tests (Bolton(1894);Wundt(1911)) is that listeners impose a rhythmical grouping pattern on sequences of acoustic stim-uli even if these are — acoustically — identical, i.e. they have the same duration, frequency, intensity and interval duration between successive events. Such stimu-lus sequences are referred to asisochronous pulse trains. In section 1.2.5we defined rhythmical structuring as a process of grouping sucessive rhythmical events into similar sequences. In this section, we propose that similarity can be achieved either by groupingan equal number of auditory eventsor bycreating similar patterns of stronger and weaker eventsin each sucessive group.

Figure 2.12: Perceptual groupings of listeners when perceiving sequences of identical stimuli, so-calledisochronous pulse trains.

When listening to a sequence of isochronous intervals, listeners tend to group them into pairs or triples, thus perceiving each second or third stimulus as more

prominent (Handel (1989); see Figure2.12). This undermines the fact that the per-ception of rhythm is very much guided by top-down expectancies or perceptual conditions that may override the physical reality. Listeners obviously feel a need to impose a structure on what they hear. The reason for this can be manifold, likely explanations are to be found in the cognitive processing of auditory events, e.g.

the temporal integration of auditory events into windows of a temporal present (cf. section 2.1.2). This means, that a chain of events may be perceived as one group due to the fact that the events contained in it belong to the same window of temporal presence. Schreuder (2006) replicated this effect for Dutch showing a different rhythmical structuring of listeners at different speaking rates — in fast speech, more syllables were grouped together than in slow speech, where listeners perceived more stressed beats. She explained this phenomenon with the limitations of the window of temporal presence. The idea that rhythmical structuring is the effect of grouping chunks of auditory percepts into windows of temporal presence receives further support by the findings that this grouping does not take place if the interval sequences are too long, too short or the pauses between them are very long (e.g. P ¨oppel(1994); Fraisse(1982)). Therefore, this type of grouping process is constrained by absolute time rather than the relative durations of the stimuli them-selves. Gestalt psychology refers to this effect as being the result of the Gestalt law ofproximity. We call this effect theprinciple of temporal presence(see Figure2.13). We

Figure 2.13: An illustration of grouping principles 1 and 2. Listeners tend to collect groups within single windows of temporal presence. In the ideal case, successive groups contain an identical num-ber of elements to which the inner clock is adjusted. In the example above, each window contains three rhythmical beats.

therefore conclude that grouping can emerge through cognitive processing strate-gies and by the result of integrating events into windows of temporal presence.

There exists a growing body of evidence that such windows of temporal presence are approximately 400-600ms long and coincide with typical durations of linguis-tic organisation in various languages and speaking styles. Fant and Kruckenberg (1996) found for Swedish rhythmical organization a reference quantum of 500ms andSchreuder(2006) showed for Dutch fast speech, that rhythmical groups contain more syllables due to window length. Thus, there seems to be a tendency to preceive a grouping of temporal events within a window of similar length and the temporal organization of speech processing tends to adjust to this window of human tempo-ral processing.

Besides integrating auditory events into groups of similar absolute duration, lis-teners obviously feel the need to make the group internal structure as similar as possible, since listeners tend to group an identical number of events, e.g. either two or three events into successive groups. Thus, one strategy of creating rhythmical harmony is called the principle of similarity in number(see Figure 2.13). This princi-ple explains the poetic technique of keeping the number of fundamental rhythmical events — beats — equal in each sucessive foot or verse. The number of fundamental beats contained in each upcoming window matches our rhythmical expectations.

However, rhythmical harmony is not only governed by grouping an identical number of events, but also by grouping events which show other types of struc-tural similarity. This may be important if the duration of subsequent events are too dissimilar to group identical numbers of them into windows of temporal inte-gration, so that other types of harmony need to be found. It has been shown that listeners tend to perceive successive events as equal in length, which leads to the phenomenon that objectively long events are perceived as shorter if preceded by short events, thus they appear more isochronous in perception. The most interest-ing aspect of this psychoacoustic findinterest-ing known astime shrinking probably is, that is can propagate across several events, e.g. an antepenultimate event can influence the perceived duration of a final event in a sequence of three (Sasaki et al. (2002), cf. Figure 2.14). Time-shrinking can be blocked, however, in the presence of strict alternation of intervals. Thus, listeners not only feel the need to impose structure on equal successive events but also try to equalize the characteristics of events within

a group, if there is a tendency towards deceleration. We calls this perceptual mech-anism theprinciple of local similarity(see Figure2.15).

However, learning may change this effect, i.e. listeners are becoming more sen-sitive to anisochrony after listening to sequences of several isochronous stimuli (Schulze(1989)). We can conclude from this that there exists a tendency to perceive sucessive events as being rather similar unless we learnt to expect a different length based on recent experiences. In such cases we are very sensitive of anisochrony (principle of local dissimilarity, see Figure2.15).

Figure 2.14: An illustration of the psychoacoustic phenomenon known as time-shrinking. A short preceding event can make a subsequent event appear as shorter. Thus, both events appear as more similar in temporal extension. The effect can propagate across several intervals, i.e. a perceptually shrunken interval may itself cause time shrinking.

Local anisochronies are probably the most well-know method of rhythmical grouping: A group is created by patterns of longer and shorter intervals, e.g. (long—

short—short) (long—short—short)or(long—short)(long—short). Rhythmical harmony is then constructed by structural similarities of long and short events across groups.

The rhythmical patterns may lead to the perception of rhythmical groups even if the group internal timing relations are not objectively identical. Handel(1992,1993) found that listeners have problems hearing differences between groups if their rela-tive timing patterns were identical on an ordinal scale, i.e. subsequent groups con-sisted of identical patterns of longer and shorter events but varying absolute relative durations. However, grouping seemed to work differently for different tempos: In slow rhythms, groups were identified based on similarities of the initial element, while in fast rhythms, groups were built based on a similarity of final elements. If a rhythm is perceived because the different groups show strong local anisochrony but a similar structural pattern, the rhythmical group is built based on theprinciple

Figure 2.15: An illustration of grouping principles 3 and 4. Listeners tend to equalize the perceived duration of rhythmical beats within each group, unless they adjusted to a particular duration in sequences of isochronous events. In such cases, they are more sensitive towards differences from the trained reference duration and perceive it as anisochrony .

of global similarity(see Figure 2.16). It is likely that our inner clock defining the du-ration of temporal presence is continually adjusted with the help of such indices of group boundaries.

Rhythmical grouping involves several mechanisms:

1. The principle of temporal presence: Groups tend to be built within identical windows of temporal integration or temporal presence. Thus, the inner clock defines the upper limit of such groups and will usually operate around 400-600ms.

2. The principle of similarity in number: Listeners prefer to have groups con-taining identical number of fundamental rhythmical events.

3. The principle of local similarity:Listeners tend to equalize the perceived du-ration of sucessive events, given constant variability, they are less sensitive to it.

4. The principle of local dissimilarity: If listeners are trained to a particular isochronous duration in successive events, they become very sensitive towards

Figure 2.16: An illustration of the grouping principle 5. Based on the perception of anisochronies, groups are built showing the same distribution of anisochronous beats. At rather slow speeds, the anisochronous beat tends to be aligned with the beginning of a group, at fast speeds with its end.

anisochronies. This means, that the more similar sequences of fundamental events are, the better is any upcoming dissimilarity perceived.

5. The principle of global similarity: If anisochronies are perceived they are a major cue to indicate the beginnings or ends of a group. Tempo defines wether anisochronies are perceived as beginnings (slow tempo) or endings (fast tempo). Global similarity is created by patterns of stronger (longer) and weaker (shorter) events in identical positions across adjacent groups.

Thus, we have two fundamentally different strategies to structure rhythmical groups: Rhythmical structure is built by the number of fundamental beats contained in them and by their pertinent pattern of relatively isochronous and anisochronous events. If a relatively harmonious sequence of groups has been processed, a lis-tener may formulate hypotheses concerning the rhythmical structure of upcoming events. These hypotheses are abstract generalizations and can be called meter. In speech, a perfectly regular meter is rare and only likely in stylized speech such as poetry, sermons or rhetorical figures. In music, however, meter is usually essential and normally obeyed by the performed rhythm within the limits of a musician’s interpretative freedom. We close this section with the following definition:

Definition 4

Meter is the abstract rhythmical generalization based on previous rhythmical expe-rience. Meter leads to hypotheses concerning internal structure of upcoming rhyth-mical groups. Depending on context, metrical hypotheses can be very precise (e.g.

when listening to a march showing a high degree of rhythmical uniformity) or rela-tively weak (e.g. when listening to spontaneous speech containing a high degree of rhythmical variability).