Friday, September 06, 2002

A cognitive take on innovation in music (esp. jazz) (potential candidate for a rewrite)

Just got back from a Cognitive Science Colloquium featuring Phil Johnson-Laird's talk "Improvisation and Innovation in Music: The Case of Jazz". The talk was a blend of ideas from cognitive science and computational theory with a smattering of music theory. Prof. Laird is from England, so I had the pleasure of listening to an interesting talk in British English and peppered with the classic clipped British wit. Although the experiments he had conducted and the results achieved were not ground-breaking or definitive, he provided interesting ways of looking at the rather abstract notion of 'creativity' from a computational perspective (I should perhaps say cognitive modelling). The case study was, of course, the improvisations of the great masters in Jazz. A jazz pianist himself, Prof. Laird also demonstrated some of the musical ideas that the talk included. His experiments proceeded from the assumption that creativity was computable to some degree. Based on this assumption, we can have three models for the process of creation: a neo-darwinian model, a neo-lamarckian model and a multi-merge model (which combines features of the first two).

The neo-darwinian model comprises two phases: the generate phase involves random variations (after all, randomness is the basic example of creativity). The outputs of this phase are subjected to an evaluate phase, which includes constraints that act as filters. Why constraints? Well, because not every random variation is interesting or innovative (in fact, a lot of them are potentially junk). What constraints are these? In the context of music, they could be accepted principles of music: metre, scales, pitch, harmony. The output is then fed back to the generative phase, as a step to potential refinement.

In the neo-lamarckian model, constraints are moved to the first (generate) phase while the second (select) phase involves a random selection from among the generated variations, thus providing the non-determinism. There is no feedback.

A multi-merge model distributes the constraints across the generate and evaluate phases, representing a blend of the previous two models.

The next section of the talk was a refresher course in tonal music and the traditional 12-bar blues structure as a foundation for jazz improvisation. Prof. Laird then introduced a multi-stage machine for the generation of these improvisations. A context-free grammar is responsible for generating the basic underlying sequences (in this case, the measures of the 12-bar blues). Context-sensitive grammars are used for insertions and substitutions over these sequences. The shift from a context-free grammar to a context-sensitive one indicates that the process of creation demands a lot from the creator, in terms of memory (of previous motifs, patterns or riffs, if you may).

My favourite aspect of the talk comprised three questions, which he attempted to answer by the end of the talk: why does most creation depend on revisions? why is it easier to criticise than to create? how do we learn to create? To answer the last question first, you learn to create by creating. A rather circular answer that, but it is perhaps better understood as the need to and importance of practice. Creative constraints are thus acquired only by practice (and lots of it too). Since critical constraints can be acquired merely by description, it is easier to criticise than to create. Since not all the creative constraints shift to the generative phase, the creator revisits the original creation to refine it, and thus most creation depends on revisions.

There were ironic moments during the talk. He mentioned parsing of parenthesized arithmetic expressions. Now this is something we had done last as part of our undergraduate programming assignments. However, as it turns out, I had just written one again, just to keep my brain cells churning. Another echo of life came when he talked about Archimedes and his famous quote (which forms an interesting portion of The 400 Blows, which we had finished watching yesterday.

So what did I get out of this talk? A better understanding of creativity. As Ulrich wrote in 1977, creativity can be regarded as the process of modifying old motifs to fit new harmonic situations (the problem of course being that someone would still have to come up with the motifs to begin with). Now, if you read that closely, it would fit what we broadly refer to as a 'cover', a 'ripoff' or a 'remix'. Take your pick. Surely, there is an element of creativity in each of these. But how do we choose the ones we censure and the ones we condemn? Is that purely a subjective decision? A matter more of personal opinions about music and ethics?

No comments:

Creative Commons License
This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivs 3.0 Unported License.