Browse Theses

Musicians made their first attempts to use computers for musical purposes about 25 years ago. A major problem that confronted them was that neither musical sound nor the standard representation of music--conventional music notation (henceforth called "CMN")--could be either input or output by existing computer systems. Since then all four aspects of the musicians' problem--sound input, sound output, CMN input, and CMN output--have been attacked, with varying degrees of success. This project is concerned with one of these aspects, namely CMN output.

Producing CMN output is most generally viewed as having three aspects: selecting the symbols to print, positioning them, i.e., deciding where to print them, and actually printing them. The positioning aspect is the core of the problem: it is essentially a question of formatting, analogous to formatting natural language text, but much more difficult. A solution to this aspect of the musicians' problem, that of automatic music formatting, would not only be very valuable to musicians, but would also be quite interesting and nontrivial from the computer scientist's viewpoint.

I argue, and give examples by major composers to show, that "fully automatic high-quality music notation" is not merely nontrivial but in general impossible without human-level intelligence. Since artificial intelligence is far from this level, one must compromise, and three ways are possible: compromise in degree of automation, in quality of output, or in generality of input. Most workers on the problem have primarily sacrificed automation. The present work, in contrast, primarily sacrifices generality of input, partly due to historical accident but partly on philosophical grounds. A major chapter of the dissertation documents my program as so far implemented. The dissertation concludes with a discussion of the central unsolved problems of automatic music formatting.