Bob Carpenter (bc10+@andrew.cmu.edu)
Mon, 11 Sep 1995 11:58:50 -0400 (EDT)
I think you responded to exactly my point. I don't see how to evaluate the CELR approach vs. the trace approach in HPSG. Both involve 'one thing' and both generate exactly the same grammar. I'm not saying that more rules are preferable -- the problem with lots of rules is that you have to find the right one, and the problem with very few rules is that they tend to apply too often. So it's a real tradeoff that will only be determined (in terms of efficiency) on a parser by parser and grammar by grammar basis. I'm all for generalizations, but what are the generalizations about? Our language as abstractly generalized by linguists, or our knowledge of language? Of course, you can argue competence/performance ala Chomsky and claim that we can have our cake and eat it too. I just claim that the psycholinguistic data isn't rich enough to decide the issue of how our 'knowledge of language' (including how to parse it, interpret it in context, etc.) is structured or deployed. - Bob Carpenter
This archive was generated by hypermail 2.0b3 on Fri Dec 18 1998 - 20:33:04 PST