A general parsing model for music and language

Abstract

Is there a general mechanism that governs the perception of phrase structure in music and language? While it is usually assumed that humans have separate faculties for music and language, this work focuses on the commonalities rather than on the differences between these modalities, aiming at finding a deeper "faculty". We present a series of data-oriented parsing (DOP) models which aim at balancing the simplest structure with the most likely structure of an input. Experiments with the Essen Folksong Collection and the Penn Treebank show that exactly the same model with the same parameter setting achieves maximum parse accuracy for both music and language. This suggests an interesting parallel between musical and linguistic processing. We show that our results outperform both the melodic component of Temperley (2001) and the musical parser of Bod (2001b).