A computational approach to modeling musical grouping structure

Abstract

A prototype cognitive model for the processing of musical grouping structures from discrete pitches is described, focusing on the level of the phrase for tonal melodies. The AGA system (“Automated Grouping Analysis”) based on this model, has been implemented as a rapid prototype in Common Lisp on an Apollo Domain AI workstation. AGA exploits two ways in which the grouping and time-span reduction components of Lerdahl & Jackendoff's (1983) theory interrelate, in terms of the normal forms of time-span reduction and the parallelism exhibited at deep reductional levels. Normal forms are implemented as the highest level rules of a grammar of chord function, to identify phrase level bound-aries, which are evaluated in terms of parallelism displayed by reductions based on well formed parse-trees according to the grammar. We conclude that processing of grouping at this level is an activity which requires a great deal of knowledge of the specific genre concerned in the form of processing heuristics, and that the listener's prior expectations could be understood within the framework of schema theory.