This dissertation uses several interrelated methods derived from corpus linguistics, statistics, and machine learning to infer a number of historically significant voice-leading schemas in a corpus of eighteenth-century Neapolitan solfeggi (exercises for voice with bass accompaniment). The goal of this work is to gain insights not only into the characteristics of this influential repertory but also into the cognitive processes by which music is learned more generally. A novel but simple approach based on the tendency of voice-leading patterns to exhibit isochrony, underlies the unsupervised learning technique. The central finding is that what the computer deems relevant tends to align with those patterns identified in historical treatises, as well as in writings by modern theorists. In this work, the concept of style is understood from a statistical perspective. This work demonstrates that musical elements that occur with each other at an above-expected frequency tend to overlap with what musically experienced listeners judge as stylistic relevant features. It is argued that we are sensitive to these probabilities and derive an understanding of music through this. That we acquire knowledge of musical style in this way invites comparison to a process identified by Saffran et. al (1996) known as statistical learning. Saffran's experiments demonstrate that even as infants, we are keenly aware of statistical properties of auditory sequences, and automatically use this information to infer the structure of our external environment. While the current work is not an attempt to model statistical learning, the current method aligns with the general concept of acquiring complex, and largely implicit knowledge, by attending to the statistical regularities contained in our auditory environment, and moreover that much of what it means to be an experienced listener might be a consequence of this type of learning. Additionally, the same method is used to examine musical patterns in a variety of corpora, including ragtime music. The purpose of these comparisons is to determine whether an algorithm designed to learn schemas in a narrowly circumscribed corpus, i.e. eighteenth century Neapolitan Solfeggi, is useful as an analytical tool for other music more broadly. Extensions and applications of the current work are explored. For instance, it was observed that the algorithm, in locating specific musical schemas, was concomitantly identifying the local meter. I offer an explanation for this emergent phenomenon, and attempt to place these findings into the larger context of cognitive theories of music. Finally, future work borne out of this study, specifically with respect to representing cognitive structures in music, is discussed.

Last modified

• 01/16/2019

Creator

• James Symons

DOI

• https://doi.org/10.21985/N2M75D

Subject

• Music

Keyword

• Galant style
• Temporal regularity
• Music cognition
• Schema theory
• Machine learning
• Structure mapping theory

Date created

• 2017-01-01

Resource type

• Dissertation

Rights statement

• In Copyright