Using Generative Grammars for Music Composition

Abstract

The use of computers in music has, to date, focused primarily on sound synthesis and signal processing. Organized research in computer music mostly has been concerned with the development of hardware and software models for the synthesis of sound and the description of instrumental and instrumental-like timbres using these models.' The application of computers in the process of music composition (that is, the generation of sound structures) has been for the most part neglected, though it is not a subject of lesser interest. The concern of researchers using computers to investigate composition is to make the rules of music, or at least of a music language, explicit. A description of a composition process by explicit and formal rules may be used to program, automate, and simulate certain aspects of the process of music composition. Some composers, such as lannis Xenakis (1971), have used computers in the process of composition as compositional aids, while others use the computer more consciously to study the way in which they compose, for example, G. M. Koenig (1970a; 1970b). Perhaps the first investigation into programming compositional rules was done by Lejaren Hiller in the 1950s when he experimented with generating compositions by making random choices within specified bounds. The first computer-composed piece is his Illiac Suite (1958). Since the early 1960s Iannis Xenakis (1971) has formalized and programmed rules of his compositional method-notably his rules of stochastic composition. He has used Markov chains to generate strings of sound textures where each texture is in turn described by probability distributions. More recently, G. M. Koenig (1970a; 1970b) has programmed serial and other composition rules to generate compositions of impressive sophistication. Though other work has doubtless been done, the brevity of the list of well-known significant work itself tells one something about the state of research in computer music composition. In all of these cases, the composer has programmed compositional rules on a computer (in a formal and explicit manner). The computer, following the defined rules, simulates the composer's rules and generates compositions (or parts thereof). The adequacy of the description of the compositional process may be evaluated from the results. A great barrier to composers and researchers attempting to formalize compositional rules-and rules of other musical activities-has been both the difficulty of making explicit these rules (i.e., what it is that people actually do when composing) and then expressing these rules to a computer.