The segmentation and ordering of an continuous sensory stream into a series of recognizable events presents one of the fundamental problems in perception . The auditory system must partition the incoming stream in a meaningful way; one which preserves relationships within the stream, but also breaks the stream into a sequence and thus allows the recognition of words, phrases or sentences.
In music, these sequential units are formalized and regularized in such a way that much of the possible ambiguity in segmentation is removed. This process of disambiguation is achieved in a variety of ways, primarily by stress or accent . Why is there a need to disambiguate the structure of such regular musical patterns? The rhythmic structure of spoken language carries content which helps the listener organize the sounds of speech into grammatical sense during the process of comprehending the meaning of the sentences . Why is there a need for prosodic elements in speech?
By studying the nature of the ambiguity in rhythmic sequences we can understand and predict the organization which will be perceived to be inherent in the timing of the sequence. This will lead to a more precise understanding of the interrelationships between stress and timing which create the unambiguous perception of segmentation of auditory streams.
A simple repeating rhythmic pattern which contains no stressed elements may be perceived as having a variety of starting points. Figure 1 shows a repeating sequence which could be perceived as having one of three potential starting points. Some starting points have a higher probablility of being perceived than others, but each of these probabilities is greater than zero.
1 0 1 1 0 1 1 0 1 1 0 1
0 1 1 0 1 1 0 1 1 0 1 1
1 1 0 1 1 0 1 1 0 1 1 0
Figure . A repeating sequence of length three has three potential starting points.
Garner and his colleagues [8,3,4] studied these types of rhythmic patterns and devised heuristics which they named the run principle and gap principle by which predictions could be made regarding the organization that would be perceived by individual subjects. The work presented here replaces Garner's heuristics with a more formal information theoretic  estimation of the probability of perceiving any starting point as a segmentation boundary. This relationship between local information content in the perceptual stream and the perception of temporal segmentation is likely to generalize to the other sensory modalities.