Grammar Induction for Musical Melodies

Universiy of Southern California, Spring 2007

in ISE575/EE675/CSCI575/PSYCH675

by Reid Swanson (2007)

Future Work

Although it is difficult to assess the quality of these parses the results do seem promising. However, there are many enhancements that could be applied to further improve the results. The quality of the parses is highly dependant on the quality of the segments. Although my segmentation algorithm seems to work most of the time there are several improvements to the way rests, pauses and null tokens are handled. It would also be interesting in testing improvements to the optimization algorithm. Even though conjugate gradient did not work for language parsing it might be worth exploring in this context. Since the introduction of the CCM there have been several interesting unsupervised dependency grammar models proposed that could also provide interesting alternatives and produce results more analogous to the work of Lerdahl's. Finally there are several other modifications to the parsing and feature model inspired by lexicalized parsing (Collins 1997) and Bare Phrase Structure (Chomsky 1995) models of grammar that I would like to implement.