Re: Virtual Concerto


Subject: Re: Virtual Concerto
From: Eliot Handelman (eliot@generation.net)
Date: Thu May 20 2004 - 20:45:44 EDT


Michael Rempel wrote:

>>>
>>>
>>db's are useful to TEST analytical tools. Whatever 'hangs together'
>>isn't something that the
>>db knows anything about, and there aren't any "tools" for
>>testing "what
>>hangs together" -- that
>>must be based either on your theory or your subjective impressions.
>>
>>
>>
>Granted. I mean that when subjective impressions are recorded, a database
>illuminates over many iterations what generally hangs together.
>

Well let's slow down here. Only late in my life did I get the idea to
just have it out with
a small handful of pieces, and it was even later before I began to see
what they had in
common. Since the moment the light bulb flickered, I've become greatly
partisan to the theory that studying music means taking ONE piece and
getting to
know it at least as well as the famous zen disciple, in the John Cage
story, whose master
asked him to describe a fish. He started to get to know it after he'd
exhausted all his
set categories and the fish was in a serious state of decomposition.
 

So though I said above that you could test theories on a database, I've
never actually
done that myself. I WAS going to do this on the essener database but
wound up, instead,
spending a few weeks each on a total of maybe three or four of these
tunes. Each
confirmed my impression that there is an amazing wealth of invention and
musical
intelligence even in simple musical artifacts, and I don't believe these
can yield to the
blunt instruments of statistical analysis David Huron wrote a paper,
for instance,
in which he tried to test the hypothesis that "many melodies take the
form of an arch."
He found that this was true. But why is sometimes not true? Are those
not good melodies?
Should the machine composer not concern itself with those exceptions?
And further,
in addition to sometimes being an arch, what other qualities need to be
featured for
a tune to make it folklorically? Which parts can dispensed with? And
besides,
what sort of musical science is this? The subjective impressions are
mostly here just vague,
half-inarticulate truisms about generalized properties ("goes up and
down,") or hopeful
theses such as "is based on harmony," I think I would make the same
criticism of
more sophisticated work that uses ML models like ripper or whatnot,
without necessarily posing
coherent and deep questions regarding the nature and structure of the music
it wishes to investigate.

I was last night looking through some papers about "e-rater." the automatic
"essay content" scoring system for GRE's.. You might like to have a look to
judge whether this is interesting or not. Easily googled.

If you're a slashdot reader, the #1 item this morning featured a fractal
algorithmic music site with java. I found it elucidating, for once, to
read through
some of the comments.

I was also flipping through Margaret Boden's book on Creativity
(explored from
an AI perspective) and reread the bits about Cohen's Aaron and
Johnson-Laird's
jazz improviser. In regards Cohen, I as always feel such envy that the
visual
problem of automatic art making is tractable, in comparison with the
almost
metaphysical problem of automatic music making. I feel I need to get a big
alligator to suspend from my ceiling to put me in mind of other hopeless
endeavors.

-- eliot



This archive was generated by hypermail 2b27 : Sat Dec 22 2007 - 01:46:01 EST