Most shamisen pieces are formed by combinations of numerous melodic patterns, and thus to understand the structure of a piece one must recognize how a melody is blocked into patterns. However, past studies of shamisen melodic patterns have encountered certain difficulties: First, the conception of the length of patterns varies widely among shamisen players and researchers, and even experienced shamisen players have only a subconscious recognition of most of the patterns. Second, although the lengths of some patterns are fixed, most vary according to context, and thus are influenced by preceding and subsequent melodic material.
In traditional musicology, studies of shamisen melodic patterns have been based on written or oral accounts given by shamisen players, an approach used by MACHIDA Kasho in “A Study of Melodic Patterns in Japanese Vocal Styles Accompanied by Shamisen, ” which presents characteristic melodic patterns found in various vocal styles. MACHIDA's study is helpful for classifying and categorizing previously recognized and fixed melodic patterns, but these patterns are only a fraction of the potential material available in shamisen music. In addition, MACHIDA's list includes patterns that range in length from one to sixty bars, and this vast diversity hinders effective blocking. Consequently, MACHIDA's study cannot provide a thorough approach for blocking a melody into combinations of patterns.
If one could estimate the approximate length of patterns before blocking, the analysis process would become more efficient. Thus, an analytical method based on information theory would be effective for blocking a piece into patterns. A scanning process would reveal recurring patterns, and subsequently ciphers containing smaller amounts of information could replace the patterns. Blocking a piece with this process would compress the original melody and thus facilitate analysis. Our study adopts the “minimum principle” from Gestalt psychology and assumes that a description of a compressed melody is a rational means by which to describe the structure of the original. Using this assumption we have blocked and compressed the shamisen part of ten nagauta pieces with shamisen-bunka notation.
We have calculated the total amount of information for describing a melody as a sum of both the amounts for registering the melodic patterns and for describing the original melody by these registered patterns. Balancing these two aspects, we replaced the melodic patterns in the encoded shamisen melodies with ciphers, thus decreasing the total amount of information necessary to describe the patterns. We used two kinds of pattern processing, blocking melodies either by patterns with fixed lengths, or by patterns with lengths that changed according to the most suitable solution for compression. Compression rates reached 67.0 percent by the former process and 59.5 percent by the latter. The latter process also enabled us to extract 1, 482 patterns from 6, 464 bars, and we then described the ten nagauta pieces as combinations of these extracted patterns. In addition, we detected cycles of one, two, four, and eight bars in the encoded melodic sequences. Consequently, this study verifies KOIZUMI Fumio's hypothesis that phrase units in Japanese traditional music are stacked in multiples of two, at least within the nagauta-shamisen genre; however, one must remember that most melodic patterns with the above cycle lengths appear as roughly-matched rather than completely-matched patterns.
Many studies have used computers to simulate human perceptions and recognitions. Similarly, our method of blocking melodies by compression based on information theory provides an approximate model for human recognition of patterns in shamisen melody, yet at the same time avoids the problems of subjective interpretations