Algorithmic Composition
In the New Century
Otto Laske
As a composer, I have been living with “algorithmic composition” since 1964, when first hearing G.M. Koenig talk about his Project One venture at Darmstadt, Germany. I did not know computers at that time but, as he put it, “I have composed algorithmically before I ever saw a computer.” He meant that, after working with Stockhausen, the “top-down design” of compositions was a natural for him, and computers emerged at the right time to help him carry out algorithmic designs. Historically, this view of algorithmic composition got lost in a plethora of actual computers, now taken for granted. However, it seems important to preserve the idea that the computer is just a tool for carrying out a particular notion of how to compose music.
The equation of “algorithmic” with “top-down” has to be understood in context. The term “top-down” here could be translated by “parametrically.” The notion is that instead of thinking of surface structure, “how the music will sound,” or about performers, you first begin designing a conceptual framework of parameters such as pitch, intensity, and others, and you think about their ranking. (You remember that in 19th century music, the ranking of parameters was largely harmony/meter=>melos as “melody”=>rhythm=>tone color.) And, in keeping with old European traditions of counterpoint, instead of punctum contra punctum, you think of “parameter contra parameter.” You envision new musics defined by flexible ways of having syntactic and sonic parameters relate to each other, perhaps even in a way that changes over the duration of a composition.
To think in terms of parameters, you need to take into account the medium in which your music is meant to sound, above all, whether it is going to be instrumental or “electroacoustic.” Not only Koenig, but Xenakis, Babbitt, and Lejaren Hiller researched in some depth what instrumental parameters lend themselves to algorithmic design, and how the limitations of instrumental parameters might be overcome by working in the electronic medium. All of these composers came up with slightly different answers as to what parameters matter, and the ranking of parameters. Their research was expanded by investigations into “computer music” in the 1950s and early 1960s (e.g., Max Matthews, Jean-Claude Rissey, and others) which gave rise to the notion of a software “orchestra” composed of “instruments” and reading/sounding a numerical “score.”
Interestingly, the “sound synthesis” approach to composition thereby created was at odds with the algorithmic approach, where the emphasis had been put on the design of the score, and where the orchestra had been seen as a medium in which to articulate the score, not as an end in itself. For this reason, in the 1970s, I introduced the notion of “score synthesis,” to keep the notion alive that the term “algorithmic” extends to both score and orchestra.
In the early 21st century, where computers are pervasively in use and a plethora of sound synthesis options exists, the idea of a counterpoint of parameters as a basis of designing scores is not common knowledge. That is not an issue for me, but the intellectual and musical loss that results from “forgetting” score synthesis is. This is so since my own compositional experience over nearly 40 years tells me that there is something to be gained musically and esthetically by nourishing ideas about “score synthesis,” especially in younger composers.
Why?
My answer to this question is a personal one. Having started out compositionally by focusing on the score and its hierarchy of parameters, I have found myself increasingly interested in the purely sonic results of score synthesis, and eventually, in overcoming the dichotomy of “score” and “orchestra” altogether. As the score increasingly became a “base score” sprouting many “algorithmic variants,” rather than a musicological end product, the question arose “why should a composition have only a single score?” Especially where, as in electronic music, instrumental limits do not exist, why should there not be different scores at different levels of a composition, from the design of tone colors (samples) to an encompassing “base score”? This is my present notion of algorithmic composition, which I am going to explain in some more detail below.
If electronic orchestras are “algorithmic” by definition (default), what is the residual meaning of “algorithmic” when applied to scores?
Its purely descriptive meaning aside, the term ‘algorithmic’ regards compositional choices made by a human linked to a machine. I called that the composer’s “alter ego” connection (1990). This connection entails that when you use a computer program to synthesize scores (thus inviting the feedback of a logic machine to your parametrical definitions of “voice,” start time, durations, and so forth, input to the machine), you have a chance to learn what you did not stipulate correctly initially. In short, you can use systematic experimentation in producing a score. However, Koenig’s idea that, in working algorithmically, you take the machine’s feedback to heart and then change your input to the machine as long as you need to, in order to reach the musical goal you are pursuing in the score, is not always easy to implement these days. In new alter egos such as Cmask you can never get the same score twice, since even identical input executed another time will give you a variant of the original score. This is limiting for two reasons. First, learning from your “mistakes” (the discrepancy between objective and result) goes away. Second, it is harder to composer incremental deviations from an unchanged base score that defines a composition’s continuity.
Let me return to the idea of overcoming the dichotomy of score and orchestra in electronic composition. (In instrumental composition, this is, alas, not a possibility, since the orchestra is fundamentally predefined, except for “special effects”). Cast in software in Csound and comparable languages, the notion that score and orchestra are separate entities is a fundamental belief of present-day composers. As a result, most composers (except e.g., James Dashow) spend more time on their orchestra than their score, coming up with very complex orchestras and rather simple scores. What is more, they are unaware of the potential of creating sonic effects predominantly by way of scores, through what I call “subscore manipulation.”
The counter-idea to the score/orchestra dichotomy is simple. In my case, its genesis has been much facilitated by Kyma, although it could equally be achieved in Csound. The idea is to use different scores on at least three levels, that of:
· producing sound samples for “orchestrating” scores (used by instruments reading samples)
· using a base score as a parametric “theme” for generating syntactic variants
· creating ‘mixin-in’ scores that are themselves variants of the base score, for enhancing the sonic complexity and contrapuntal variety of sounding results.
When adopting this approach, the composer is working with at least 2 if not 3 orchestras, and with at least 2 if not 3 scores from the start. The notion of “the orchestra” and “the score” has been left behind. Equal attention is paid to formulating the syntactic “score” and the sonic “orchestra” part of a composition. Here, it is the mobility of the score in its different manifestations that helps along the multiplicity of the orchestra. Below is an outline of a course meant to teach this approach to algorithmic composition.