Thursday, December 06, 2007
The chicken is the egg (or is it the other way round?)
With the release of the ETC source, the basic problem of an architecture for generating poetic texts is a solved problem. So then how come the texts still fall a bit short as examples of poetic speech?
Because....
It's not just the generational software that makes the poem, but the grammar(s) the software uses to guide its semantic and syntactic choices.
As I've said many times on this blog, Etc3 uses an LTAG approach. There are all kinds of advantages to using LTAG, not the least of which is how easy it lends itself to description via Xml. The trick is what sorts of grammatical units are poetic.
There is a Sylvia Plath line with the phrase "Between this wish and that wish." (And again, I don't care for her work, but it has readily observable poetic markers that serve well as the basis for experimentation.) Now that phrase poses interesting technical challenges. Between and and are easy. We just treat those as lexical nodes. It's this wish and that wish that are problematic. To define these in a tree, we need a repetition node, which at first looks easy, but has a couple of boobie traps. Basically the repetition node searches down from its root for a node with a target ID and then requests of that node its terminal string. The repetition node cares not a lick for the grammar of that node, just its surface. So far so good. But this and that aren't so nonchalant. They have to agree with their head nouns. But the second doesn't link to anything, because its associated node is all and only surface. And they have to be opposites. The solution is to link the demonstratives so that the second thinks its determining the first wish, but change the reference from near to far. Whew.....
Lots of fun.
But the utility of the phrase is limited. How many times can you say between "this X and that X" before the phrase becomes repetitive. Not often. So as much fun as it might have been to code and as useful as it was for thinking about poetic text generation design, we're probably going to trash it.
The thing is, you can never know the affect of a construction until you see its realization within the context of an individual piece and the context of a body of work.
Keep those cards and letters comin', folks!
# posted by Jim @ 11:22 AM 
