Out of Egypt:

the shape of music, part I (toward the deep logic of motion)

The following two entries, on music, are admittedly ambitious and, at points, obscure (especially since they combine my interest in music with my casual, and possibly erroneous, understanding of programming). I can only hope that the links I provide will make up for the inadequacy of my explanation. As for the latter part, my own vision is still forming, so I can at best hope to give you a glimpse of it. If I can excite one person's interest - if I can bring them to see at least half of what I see - then I will be happy. For I think these are the kind of things that will best work themselves out in intelligent dialogue. And so, I begin:

There is a grammar - a deep logic of motion [cross out "deep structure" - I'd never heard of Schenker, and I'm not talking about melody/harmony and "important notes"] - to music which I'm just beginning to understand. I think that conventional notation has actually hindered my learning in this respect, since it does a better job of indicating what the individual pitches are that make up a piece than it does at showing phrase rhythm or the other, more global, characteristics of the piece. Notations like "D.C. al capo", repeat signs, etc. have always seemed both confusing and inadequate to their purpose.

Yesterday I woke up with a strange and intense desire to download a program which generates intelligent-sounding, "human-like" music. The closest I came, for the Mac at least, was something called Lexikon-Sonate, an endless piano "piece" described by its creator, Karlheinz Essl, as a "musical hypertext." Now, for our purposes, the most interesting thing about this program is its sense of pacing - the way it transitions between the ten or so different styles of playing which have been programmed into it. These stylistic "gestures," separated by fermatas, could be called the cellular units that make up a Lexikon-Sonate performance. The order which arises from their differentiation is what distinguishes Lexikon-Sonate's output from other attempts at computer-generated music.

As exciting as playing with Lexikon-Sonate was for me, however, it still wasn't quite what I was looking for. I had a more specific search query in mind, although I wasn't getting many relevant results for it. I wanted a program which could generate infinite performances of Terry Riley's "In C," a minimalist work which I consider ideally suited for computer performance.

The Riley-O-Matic, by Daniel Iglesia, was the only thing I found that approximated my ideal. Unfortunately, it didn't produce "In C" exactly (because it wasn't programmed with Riley's cells or the rules of the piece). More importantly, it wasn't available for download.

Disappointed, yet intellectually stirred, I continued my search. Exploring the links from a site called Generator.X, I found a fascinating blog entry called "The Visual Context of Music", which showed various notation systems which are strikingly different than the conventional one, whether because of their antiquity or their modernity. Seeing these, I was reminded once again how much mental models (and our written representations of them) can constrain (or liberate) our thought. The scores by Takemitsu, Guy, Andrews, and Crumb, especially, highlight two properties of musical structure which conventional notation obscures: music's cyclical nature and its decomposability into basic forms, the atoms of melody which we call phrases. While a lead sheet, with its reduction of a song to chords and melody, hints at these, it is not a total representation of a work, as the scores presented at BiblOdyssey are.

From BiblOdyssey, I followed the link to a site called The Shape of Song, where MIDIs are analyzed and visually re-presented (cf. Visual Complexity) in an attempt to "answer the seemingly paradoxical question: What does music look like?" Understanding their method, which presents a musical piece "as a sequence of translucent arches," each of which "connects two repeated, identical passages of a composition," led me closer to the insight I was seeking.

The only flaw I saw in their algorithm was the indiscriminate way in which it was applied. Repetition in the macrostructure and microstructure of a piece was shown in the same way, with no way to distinguish between the two other than the width of the arches. In many pieces, in fact, it became difficult to tell what the more important sections of a piece (as identified by repetition) because the smaller (and perhaps coincidental) points of repetition were so numerous that they obscured the deep structure, like a forest of bridges.

What we need, I decided, is a dynamic representational system - one with expandable levels of detail. Now that hypertext has become the dominant medium of our age, and the network its dominant metaphor, we should no longer use a system of musical notation limited by the linearity of the printed word. Though music, as performed, is linear - it exists as a movement through time - as conceived, composed, and skillfully interpreted, it exists as a totality. Along with the horizontal structure of melody and the vertical structure of harmony, we must recognize the "third dimension" of overall form - the balanced pattern of repetition and variation which gives music both coherence and interest.

Furthermore, only when our mental model of music (as shaped by its external realization) has been enriched by an understanding of the cross-temporal pull of musical units (how one phrase calls another to remembrance, etc.) can we begin using the categories of our mental model to make an artificial intelligence sophisticated enough to create music that sounds like it was composed by a human being. Current computer music is either too repetitious (and thus boring) or too random and capricious (thus irritating and incomprehensible). Not only in its understanding of harmony (as governed by the principles of consonance and dissonance), but in its understanding of form, must the computer be trained in the balance between order and freedom.

Through listening to "In C" and studying its score, as well as looking at the interfaces for the programming languages Max/MSP and Squeak, I have come up with an interface concept for a music composition/generation environment which both: 1) possesses structures rich enough to represent nearly all music (in the "third dimension" of form as well as the horizontal and vertical dimensions - melody and harmony) and 2) can serve as a platform for both human and computer composition and improvisation. I believe that, for the foreseeable future at least, "computer-generated music" will involve a fair amount of behind-the-scenes work by humans who select melodic and harmonic material and rules which will create interesting pieces. That said, a computer-aided composition environment with a sophisticated representation of musical structure, as well as the ability to generate an infinite number of pieces from a given framework, will be a great help to human creativity. Insofar as music is mathematical (more specifically, algorithmic), the vast calculating power of computers can help accelerate our exploration of the space in which interesting music is found.

Man, Evan, you sound like a modern day Schenkar...German musicology would be so proud of you. :)

I also find it somewhat amusing when people try to "scientize" music notation, because music notation is, at heart, nothing more than a metaphor. Of course, metaphor can certainly rely on technological models for communication, again, the degree of usefulness lies in how intuitive Analogizer is (because if the Analogized is easier to understand than the Analogizer, then what is the point of analogy at all?).

I have to admit I haven't fully read everything here. But since when has partial understanding prevented me from commenting?

Posted by: funke at May 29, 2007 3:20 PM

Hm, I also didn't fully read my comments either. Typos abound. Sorry about that.

Posted by: funke at May 29, 2007 3:23 PM

PS Have you tried the new iLikes feature on facebook? I think that deep structure theory must somehow be responsible for some of the "sounds like" options that come up. Honestly, the Paper Street Saints are the not first group that comes to mind as similar to Brahmns. But then, they are a metal group, and I've always felt that metal and the Romantic composers had some definite affinities...

Posted by: funke at May 29, 2007 3:39 PM

Shenker *shudder*

Have you heard of Pandora.com? Also one of those scientizing things...I find it unsatisfying for me. *shrug*

Posted by: Jeannette at May 29, 2007 9:43 PM

To clarify, Jeannette, I wasn't talking about Shenker, since I know nothing about him. I don't think Pandora or iLike is really like what I'm talking about. Really, what I'm talking about is a combination of the Lexikon-Sonate program I linked and the rules of "In C," generalized to account for most tonal music.

Posted by: Evan Donovan at May 29, 2007 9:57 PM

I was the one who mentioned Schenker. I think a lot of music analysis reminds me of Schenker. Maybe there is some deeper structure to musicology...or perhaps I was just scarred for life having to draw diagram after diagram...

Posted by: funke at May 30, 2007 12:19 AM