Abstract:

History in composition and analysis have shown that composers using the same patterns in structure and harmony get different results depending on the way these patterns are resolved. In terms of musical analysis, a particular piece can be described by a sequence of states and transitions between states representing the personal criteria that each composer pursues when solving a musical structure. A first-order learning system, called Pal, is used to learn transition criteria for counterpoint analysis, in the form of Horn clauses from pairs of musical states (given as sets of notes) and general purpose musical knowledge. It is shown how the rules learned by Pal can be used for musical analysis of simple two-voice counterpoint pieces. Similarly, a counterpoint voice can be generated from a single voice (cantus firmus) using the learned rules. Conclusions and future research directions are given.