A large portion of my interests lately has centered around the proposition that syntactic theory has been overprivileging semantic argumentation. I’ve talked a lot, and in various forums, about why I think this is bad for syntax. Here, I’d like to say a bit about why I think this is bad for semantics.
Let us stipulate that a central goal of any semantic theory is to be compositional. The slogan here is often something along the lines of, “How the meaning of a complex expression is related to the meanings of its parts.” But I think this particular formulation actually obscures a crucial point. For anyone who is not a flat-Earthist about syntax (à la “Sentences are just flat sequences of words!”), a compositional semantics is not about relating the meaning of a complex expression (directly) to the meanings of its atoms. Instead, it’s about relating the meaning of every constituent to the meanings of its immediate daughters.
It follows from this that a compositional semantic theory is only as good as the syntactic theory it is hooked up to. A syntax developed completely in the service of semantics – i.e., using argumentation from meaning as its principle guiding light – risks rendering the entire project of compositional semantics circular. If you have a syntactic structure crafted to capture (mainly) generalizations about meaning, then achieving compositionality in the semantic analysis is really no achievement at all.
Let’s clarify: computing the meaning of an utterance from the meanings of its atoms in one way or another is already no small feat. But such a computation can lay claim to being a compositional model of natural language semantics only to the extent that it composes those atoms in the same manner as the syntax (of natural language) does. Therefore, once we base our theory of the latter – how the atoms compose syntactically – on meaning, we compromise the goal of truly building a compositional natural language semantics.
Fortunately, there is a fairly straightforward way to break out of this circle: base your syntactic theory (primarily) on string-acceptability, not on semantic interpretation. If your semantics is hooked up to a syntactic theory of this kind, then compositionality becomes a meaningful achievement.