Creating a nested melodic representation: competition and cooperation among bottom-up and top-down Gestalt principles

A set of principles (based on Gestalt theory) governing how we group notes into meaningful groups has been widely accepted in the literature. Based on these principles, many divergent theories of melodic segmentation and representation have been proposed. However, these theories have not succeeded in achieving a comprehensive and verifiable representation of melody. This is largely due to the fact that multiple competing segmenting factors produce, for any single melody, a large number of possible segmentations and therefore representations. Here a model is proposed, which incorporates widely accepted principles of segmentation. These rules govern three types of factors: (1) changes in proximity (for producing disjunctive segmentation), (2) changes in overall contour and intervallic texture and (3) patterns and periodicity that create parallelism among segments. Because of the nature of the segmentation rules, these same rules establish the attributes of the groups they produce. Based on original research in Singer 2004, principles for establishing preferences among competing rules are formulated in order to create a few preferred representations for approximately 1,000 monophonic folksongs.