以前，Kai von Fintel の Semantics etc. で紹介されていたのを思い出して，改めて探しました：
Barbara Partee,“Reflections of a Formal Semanticist as of Feb 2005″
And MIT has remained the dominant institution in syntax ever since the 1960’s. It’s also part of the social dynamics of the field that Chomsky values theoretical argumentation above all else, and the sense of some things being “Interesting with a capital I” that I picked up on in graduate school hasn’t changed: I had to learn not to say too publicly that my reason for being in this field is just that I think it’s fun, and I had to learn to slightly self-censor my choice of problems to work on to be able to argue that there was some ulterior theoretical importance to such-and-such a problem. Actually, I suppose I have internalized that last part by now. It’s hard to tell whether it has become part of my own internal make-up, or whether I have simply learned to look for theoretical relevance, but I certainly do look for it now, both for my own sake in writing grant proposals, and in advising students in their choice of topics and in writing referee reports for conference abstracts, journal articles, etc.
Formalization is an excellent thing in moderation. When there’s too little, claims tend to be fuzzy and untestable, and argumentation can’t help but be somewhat sloppy. But there can be such a thing as too much formalization, or premature formalization, or formalization that has too little relation to anything empirical. So even a formal semanticist or formal syntactician or formal anything else shouldn’t hesitate to write things down and share them with colleagues when they are still in an informal state; often it requires looking at things from many points of view before a good path to formalization emerges.
Why has Chomsky been so resistant to compositionality? It’s such a natural hypothesis. Semantics has to have some kind of syntax to build on – we can’t put the meanings of the parts together without some notion of what the “parts” are, and that’s syntax. So it would seem an unpleasant design flaw if natural languages had to have two different syntaxes, one just to describe what sentences there are and their structures, and another to provide the syntax that the semantics needs. Of course things aren’t always simple, and there can be mismatches of various sorts around the edges, as there seem to be between any two components. But Chomsky’s resistance to the idea of compositionality has seemed much deeper than any rational arguments could explain, as far as I could see. My own speculation – and this is pure speculation – is that it may stem from the same source as his deep resistance to any kind of functional explanations in syntax, which also seem deeper than a simple rational skepticism. Chomsky has often pointed to some of the oddest principles in syntax (e.g. “disjunctive ordering”) as evidence for an innate special-purpose language faculty, citing as a plausibility argument that something as syntax-specific as disjunctive ordering couldn’t very well follow from any general cognitive principles, and certainly couldn’t be learned, so it must be language-specific and it must be innate. It seems that any kind of outside explanation of anything in syntax is unwelcome because it might weaken the argument for this special innate faculty. Semantics may seem too close to general cognitive faculties for comfort, and compositionality requires a homomorphism between semantics and syntax, so compositionality would seem to weaken the important thesis of autonomy of syntax. In fact it doesn’t weaken descriptive autonomy at all: syntax can be described independently of semantics but not vice versa. What it weakens is explanatory autonomy: compositionality makes it natural for the language acquisition mechanism to work on acquiring syntax and semantics in parallel, with any evidence for how the semantics of one’s language works providing indirect evidence for how its syntax is structured and vice versa. I have not done any serious research in this area and haven’t written anything about it in any scholarly publications, but in this “opinions” section I will assert that I think you can learn a language a lot better by watching TV than by only listening to the radio, and that that’s one of many bits of evidence that we learn syntax and semantics together. Another anecdotal bit is the constant folk etymologizing of idioms, suggesting a very strong drive to understand language compositionally. More serious work on the interaction of syntax and semantics in acquisition has been carried out by Lila Gleitman and her students, and some by Stephen Pinker.