Categorization of Musical Patterns by Self-Organizing Neuronlike Networks

Publisher Website
Google Scholar
Abstract
Simulations of self-organizing neuronlike networks are used to demonstrate how untrained listeners might be able to sort their perceptions of dozens of diverse musical features into stable, meaningful schemata. A presentation is first made of the salient characteristics of such networks, especially the adaptive- resonance-theory (ART) networks proposed by Stephen Grossberg. Then a discussion follows of how a computer simulation of a four-level ART network—a simulation dubbed L'ART pour l'art—independently categorized musical events in Mozart's six earliest compositions. The ability of the network to abstract significant voiceleading combinations from these pieces (and in fact to detect a possible error in the New Mozart Edition) suggests that this approach holds promise for the study of how ordinary listeners process music's multidimensional complexity. In addition, the categorizations produced by the network are suggestive of alternative conceptualizations of music's hierarchical structure.