We do this (in a highly pruned fashion, lest the data structure
involved - a directed, weighted graph - exceed the size of the
universe) and then use artificial intelligence techniques to "learn"
and enforce the "grammar" of the corpus involved.
Chaos is used in our scheme to do the larger-scale shuffles (cf.,
Merce Cunningham's randomization of chunks of a dance) via a
symbolic-dynamics scheme that maps the movement progression onto a
chaotic attractor, then using new trajectories on that attractor to
generate variations. The fixed attractor geometry guarantees that the
variation resembles the original in several mathmatically precise (and
stylistically appealing!) senses; sensitive dependence guarantees that
it is different.
The problem is that this chaotic shuffling introduces abrupt
transitions - as Cunningham's scheme did - and we needed some
dynamics-faithful way to smooth them. We could, of course, have used
F=ma (as does Jessica Hodgins at Georgia Tech; see
do this, but the machine learning/computational linguistics stuff
seemed like fun.
> This, by the way, is exactly how chaotic "motion" through quantum
> processes proceeds.
Hm. Have you read Joe Ford's papers?
-- ================================================================+========== + Liz Bradley Assistant Professor \ + + Department of Computer Science \ + + Internet: firstname.lastname@example.org O )) + + Voice: (303) 492-5355/ Fax: (303) 492-2844 / + + URL: http://www.cs.colorado.edu/~lizb/Home.html / + + USMail: University of Colorado Campus Box 430 ((O + + Boulder CO 80309-0430 + + ===> NO SPAM, NO JUNKMAIL <=== + + email spam is ILLEGAL, per USC Title 47, section 227 p (b)(1)(C) + + See http://www.law.cornell.edu/uscode/47/227.html for details + ===========================================================================