A self-learning musical grammar, or 'associative memory of the second kind'

A context-sensitive generative grammar that learns its production rules automatically from examples and optimizes the length of context for each individual production rule on the basis of conflicts occurring in the source material is described. The grammar has been applied to the generation of new melodic passages and counterpoint according to a certain style. This report describes some of the principal ideas and the work that was done. Music produced by this method generally sounds smooth, continuous, and pleasant.<<ETX>>