Learning production probabilities for musical grammars

While there is a growing body of work proposing grammars for music, there is little work testing analytical grammars in a generative setting. We explore the process of learning production probabilities for musical grammars from musical corpora and test the results using Kulitta, a recently developed framework for automated composition. To do this, we extend a well-known algorithm for learning production probabilities for context-free grammars (CFGs) to support various musical CFGs as well as an additional category of grammars called probabilistic temporal graph grammars.