Abstract

Stephen Schiffer has argued that natural languages do not have compositional semantics. But it has been widely held that compositional semantics is required in order to explain how it is possible that we have the linguistic capacities that we do. In particular, our use of natural languages is productive in the sense that there are indefinitely many sentences that we have never heard or considered before, but which we are nonetheless capable of understanding. How is this possible? Compositionality evidently supplies a clear answer to that question, because it guarantees that there is some way of determining the meaning of each sentence of the language from a fixed and finite base of semantic value assignments. This poses a serious challenge to Schiffer's negative thesis. Schiffer proposes to answer this challenge in a way that will also provide a solution to the language-relation problem. This is the problem of specifying what relation must obtain between a population P and a language L in order for L to be a language of P. Schiffer's strategy is to reduce the problem for public languages to that of specifying the language-relation for languages of thought-specifying what it is to think in a language. I will show in a precise way that Schiffer has neither met the productivity challenge nor solved the language-relation problem. Using Schiffer's characterization of what it is to think in a language, I show that if an agent thinks in some language L, then there is an infinity of languages that the agent also thinks in with the very same sentence tokens, but with arbitrarily different meanings. Thus, Schiffer has clearly not given a sufficient condition for an agent to think in a language, and Schiffer cannot do with less than a sufficient condition. Moreover, I will argue that Schiffer cannot avail himself of various attempts in the literature to address similar problems.