Grammar Induction for Musical Melodies

Universiy of Southern California, Spring 2007

in ISE575/EE675/CSCI575/PSYCH675

by Reid Swanson (2007)

Language & Music

Despite the fact that language is one of the most complex cognitive abilities all humans have the innate ability to learn to use it. Without much knowledge of how our language faculty works, nor without any special linguistic training we communicate efficiently and effortlessly all the time. Much like language, the ability to appreciate, recognize and even produce music also seem to be innate abilities shared by all humans. Striking in itself, it is also likely that language and music share more than being amazing innate abilities. Current neuroscience research suggests that both may in fact share certain syntactic processing regions of the brain (Patel 2003). Although each system has a unique symbolic and cognitive representation that do not easily map from one to another there is evidence that both systems rely on hierarchical organization for processing. Theories such as Gibbson's Dependency Locality (1998) theory or Hawkins' performance theories of constituency (1994) provide convincing empirical evidence of the syntactic structures of language. On the other side Lerdahl's Tonal Pitch Space theory (2001) offers some evidence that hierarchical organizations of pitches can account for phenomena that cannot easily be explained by linear processing. These deep similarities between language suggest that techniques used to analyze and process linguistic data may also be successful in the analysis of musical data. It is the goal of this work to investigate the usefulness of applying unsupervised parsing techniques from the computational linguistics community for the analysis of musical data.