Discovering characteristic patterns from collections of classical Japanese poems

WAKA is a form of traditional Japanese poetry with a 1300-year history. In this paper, we attempt to discover characteristics common to a collection of WAKA poems. As a schema for characteristics, we use regular patterns where the constant parts are limited to sequences of auxiliary verbs and postpositional particles. We call such patterns FUSHI. The problem is to automate the process of finding significant FUSHI patterns that characterize the poems. Solving this problem requires a reliable significance measure for the patterns. Brazma et al. (1996) proposed such a measure according to the MDL principle. Using this method, we report successful results in finding patterns from five anthologies. Some of the results are quite stimulating, and we hope that they will lead to new discoveries.