Machine Learning of Jazz Grammars

Melodies It is reasonable to regard a sequence of terminal symbols in the grammar as being an abstract melody, in the sense that multiple melodies will fit the sequence when the note categories are instantiated to corresponding pitches. Another advantage of such melodic abstractions is that they can be instantiated over any chord progression, even for chords of different quality, such as major, minor, diminished, dominant, etc. Although individual note categories can be used to generate somewhat convincing jazz melodies, in order to capture specific styles it is necessary to introduce one or more mechanisms to provide greater coherence among individual notes. Thus we extend the individual note categories with “macros” that can capture sequences of notes in certain patterns. The current work focuses on a single macro concept, called a slope. Each slope has two numeric parameters, followed by a sequence of one or more terminal symbols. The numeric parameters indicate the minimum and maximum rise between successive notes in the sequence. Negative numbers indicate fall rather than rise. In the grammar rules, slopes are treated as terminal symbols appearing in the consequent of a production. In generating a melody, the terminal symbols inside a slope are converted to specific notes, as before. Figure 2. Two example