Monday, April 17, 2006
Why is shit music?
I had multiple motivations for building ETC. One of the more obvious was to find out how readers would respond to such texts. It turns out that they respond very differently when they know the text came from the machine from how they do when they don’t. Tonight I’ll talk about the responses when they know. I’ll save when they don’t for another post.
When readers know the poetry is coming from the machine, I get emails with phrases such as “Erica freaks me out” and “This is very creepy.” Not precise terms of literary analysis to be sure, but pretty much in the vein of this polemic. But why freaky and creepy? The reader knows that the software was built to simulate meaning and ought not be that surprised when it does.
Certainly there is some measure of what Dr. Johnson meant when he noted that the wonder of preaching women and walking dogs is not in that they do those things well, but that they do them at all—machine poetry as freak show. And some of the reaction probably stems from too weak a resistance to anthropomorphizing the machine—especially from skeptical readers, the ones who expect nothing but don't get nothing.
But there’s something else at work, wrapped around (or imbedded in) the fact that since these poems are “written” completely devoid of authorial intent, any meaning readers find has to be meaning they’ve themselves constructed.
I wrote the very first version of this software in Python, an interpreted and therefore slow language. The source text was Jane Austen’s novels. It took about 30 minutes to initialize and after that about 10 minutes to compose a one-line poem. It was incapable of writing more than one poem at a time. Really tedious work. I was trying to demonstrate that the machine could compose sentences with different structures. I had to devise the simplest possible test because of the software’s performance problems. So I developed code to write either declarative or imperative sentences. This was easy--all I had to do to code for the imperative version was to have the code generate a declarative sentence without a subject. It took four months to write, four months without a single successful test and then after all that time and effort, out came the very first successfully composed poem:
Imagine in this age.
Now I was freaked out. In its coming-of-age test, the machine had spoken to me about the topic of the code itself!
But of course it hadn’t. This poem was meaningful to me, not because of what it denotatively demanded of me, but because of my need for the machine to make sense, the affective demand all poetry makes on readers. I wanted my machine to mean—and I would make it mean, regardless of which set of transductions the software computed and whatever its grammar’s terminals might be.
The machine continues to speak to me. During one recent test in which I was trying to improve ETC’s facility with determiners, it spun out this line: Why is shit music?
Why indeed?
# posted by Jim @ 10:36 PM 
