Book Reviews: Machine Translation Today: The State of the Ar

Nevertheless, he argues that the evidence for a compositional truth-theoretic semantics is pretty glib, because one can give a correct model of language understanding and productivity without assuming a compositional truth-theoretic semantics. In the seventh chapter of the book he suggests a counterexample to the compositional view about language understanding. He describes a possible world in which a human computer, Harvey, understands a language E, as complex as English. The author analyzes Harvey's comprehension of some sentences and he draws a picture in which heuristic tools, such as the 'conceptual roles' of some expressions of Harvey's neural language of thought, allow him to show that "there could be a correct psychological model of a person's language processing that does not presuppose a compositional semantics for the mastered language" (p. 205). But one might object that if compositional semantics is not needed in order to account for language understanding, it is needed in order to explain what Schiffer calls "the platitude", i.e., the idea that the meanings of sentences are determined by their syntax and the meanings of their words. In Chapter 8, Schiffer argues against this objection by introducing other heuristic notions, such as the "saying-potential" and the "processing role" of linguistic expression. In the sentence "Michel believes that his car was stolen", the word "believes" has a processing role that determines its saying potential each time the word is used. "Believes" is a semantic primitive, "but it is not a semantic primitive in any sense appropriate to a compositional semantics, for no base axiom, no satisfaction clause, can be written for 'believes' that could take its place in a true truth theory for English" (p. 216). Schiffer does not want to deny all the aspects of semantic compositionality, nor that natural languages contain truth-affecting iterative devices (p. 208). What he denies is the relevance of truth-theoretic semantics to an account of language understanding and productivity. He denies that something like analytic philosophy is possible, but in the meantime he suggests that the compositionality of natural language should be explained via cognitive models of linguistic behaviour. At the very end of his book, he seems to set up an alliance with cognitive science in order to explain facts about language which do need explanation. They are facts about language comprehension, about the ways in which we store, represent, and process information. But, Schiffer says, these "are not philosophical questions (although the skills of the philosopher would be relevant to answering them)" (p. 271).