Bob Carpenter (bc10+@andrew.cmu.edu)
Thu, 7 Sep 1995 13:00:50 -0400 (EDT)
Let me clarify a couple of points. The remark about partial evaluation is not new. This is just what I was trying to point out. It keeps getting proposed as a new idea in the GPSG/HPSG literature, though. I just don't see what the evidence for phrase sturcture is supposed to be other than a theory internal notion of capturing generalizations. Similar generalizations can be captured in Lambek-style categorial grammars which have no corresponding notion of phrase structure, seem to capture the same kind of generalizations, etc. So how am I to choose between the two? It seems to me to come down to aesthetics. I'm perfectly happy to accept grammaticality judgements of speakers as evidence, but this is not what we're trying to model in theoretical linguistics. No speaker can accept multiple center embeddings, yet our grammar models them. This is because our grammars are generalizations of some kind (again, this is not a claim I'm making -- it's been made for ages). Further, assuming there's no psycholinguistic evidence for or against traces (this is a contentious point, but I'm not in a position to judge), how do we decide between the empty category account of Chapters 1--8 in the new HPSG book, and the lexical-rule-based account of Chapter 9? I can't believe anyone would argue that the lexical rule account captures more generalizations. Now, are we trying to model humans knowledge of language, or an idealized abstract system? If it's the former, then of course we have to look to psychological evidence (of all kinds). This is because just looking at the extensional side of things doesn't decide the issue (until we employ aesthetic judgements about what's more general). The problem about human knowledge is that if we use a bunch of rules in tandem on a regular basis, we start to partially evaluate them to form a single rule (the notion of chunking in psychology). Of course, the original multiple rules give us more generalization, but they're harder to use in real time. I was just trying to point out that in the case of empty categories, there seems no evidence that I'm aware of to decide the issue either way. - Bob Carpenter PS Bob: do you have a reference for the Quine paper 'Methodological reflections on current linguistic theory'? And where does Chomsky argue that the availability of extensionally identical grammars is of no great importance? Is it because he thinks we can pick out the 'right' one (on what basis?) or because he thinks if we get the extensional grammar right, which one we choose doesn't matter? (I would assume the former, but this is what I seriously doubt.)
This archive was generated by hypermail 2.0b3 on Fri Dec 18 1998 - 20:33:03 PST