DISCLAIMER: I don’t think the following thoughts are particularly “revolutionary” – i.e., I imagine other people have thought (and maybe said) the same things before. But for what it’s worth, I’m a syntactician who has been in linguistics for years and I’ve only recently come to realize these things; so perhaps there are others like me out there who might still benefit from reading this.
A while ago, on facebook, I was talking to Heidi Harley about my unease with proposals that resort to statements like “XP undergoes syntactic movement because it is of the wrong semantic type to compose in situ.” I mentioned I felt that there was something intuitively wrong with this idea, because when a movement requirement is not obeyed, the feeling I get is not that of coercion, i.e., trying to interpret something in the wrong semantic context. It is a crisply different feeling, of ill-formedness and, sometimes, outright word-saladness.
The original examples of coercion I came up with in the discussion involved things like the “universal grinder” (a bit of banana) and the “universal packager” (three waters). Seth Cable then corrected me that these examples had nothing to do with semantic type; banana and water are both predicates of individuals and, thus, equivalent as far as semantic types go. Some further back-and-forth followed and, eventually, Seth explained that “type-clash results in ungrammaticality, not a feeling of coercion.”
I think we can come up with examples that aren’t susceptible to the pitfall of the banana/water cases. For example, trying to interpret an individual as a verb, as in, say, He tried to Chomsky me to death – which, again, does not result in the same kind of sharp unacceptability that syntactic violations give rise to, and instead gives rise to something like a feeling of coercion. This, even though proper names and verbs do not share the same semantic type.
Even if such examples didn’t exist, though, I think there would still be something very wrong with the idea that type-clash results in ungrammaticality. Here’s why. If type theory is a purely formal mechanism for determining which constituents can and cannot stand in a sisterhood relation – in this case, at LF – it should worry us (a lot!) that we, um, already have a mechanism for that. It’s called “syntax” and we already have a theory of it.
This is worse than a mere duplication problem (which would already be bad enough, on its own). If you examine type theory qua theory of syntax, it amounts to a version of Categorial Grammar. And here’s the thing: Categorial Grammar is a really bad theory of syntax! (The reasons are manifold, and I really, really don’t feel like turning this post into a “what’s wrong with Categorial Grammar” post. That sounds tedious and boring. Suffice to say that it cannot handle but the very simplest case- and agreement-patterns.) Don’t you think that if we lived in a world where something like type theory underpinned constraints on syntactic sisterhood, the opposite would be the case? Wouldn’t we expect that Categorial Grammar would be a good (if not the best) theory of syntax?
Essentially, then, any theory that subscribes to “type-clash results in ungrammaticality” (and its offspring, “XP undergoes syntactic movement because it is of the wrong semantic type to compose in situ”) is a theory where there are two syntaxes running in parallel: one, the actual thing people like me would call “syntax”; the other, a syntax based on what is manifestly the wrong theory of natural language syntax (Categorial Grammar). This shouldn’t be anyone’s idea of a good overall theory of grammar.