Discourse Analysis of Lyric and Lyric-Based Classification of Music

Lyrics play an important role in the semantics and the structure of many pieces of music. However, while many existing lyric analysis systems consider each sentence of a given set of lyrics separately, lyrics are more naturally understood as multi-sentence units, where the relations between sentences is a key factor. Here we describe a series of experiments using discourse-based features, which describe the relations between different sentences within a set of lyrics, for several common Music Information Retrieval tasks. We first investigate genre recognition and present evidence that incorporating discourse features allow for more accurate genre classification than singlesentence lyric features do. Similarly, we examine the problem of release date estimation by passing features to classifiers to determine the release period of a particular song, and again determine that an assistance from discoursebased features allow for superior classification relative to single-sentence lyric features alone. These results suggest that discourse-based features are potentially useful for Music Information Retrieval tasks.