Modularity and Computation in Semantic Theory

Various features of natural language seem like they require us to revise the basic architecture of semantic theory. Some important examples are variable expressions (like free and bound pronouns), alternative-denoting expressions (like indefinites and “wh”-words), focused expressions (as in “I only have MYSELF to blame”), supplements (as in “Dave, the chief of police, arrested Clayton”), and quantifiers (as in the two readings of “a doctor examined every patient”). Each of these phenomena challenges in its own way the Fregean architecture of our baseline compositional semantic theories, in which complex meanings are built by applying functions to arguments.

What form should such revisions take? Can they be handled in a uniform way? In this talk, I’ll describe a general way of approaching these questions, using tools developed by functional programmers for structuring computer programs. Drawing on work by Chung-chieh Shan and others, I’ll suggest that the phenomena described above have well-studied analogs in programming contexts, and that the tools and theory used by functional programmers and computer scientists for reasoning about so-called “impure” features of programming languages can be used to ground and enrich linguistic theory. Specifically, I’ll touch on applicative functors and monads, showing how they allow us to modularly extend our baseline Fregean semantic theories — ones oriented around functions and arguments — in order to handle more and more data. I’ll argue that this strategy opens the door to new ways of theorizing about these phenomena and their interactions, with empirical and conceptual benefits.