Linguistic Side Effects

Harvard University, Cambridge MA

Investigators

Abstract

With National Science Foundation support, Dr. Stuart Shieber will conduct three years of interdisciplinary research on noncompositionality in natural language, where the meaning of a phrase seems to depend on more than the meanings of its parts. This research is based on an analogy between computational side effects in programming languages and apparent noncompositional phenomena in natural languages. The investigators hypothesize that techniques developed in the programming language community to specify side effects, such as various "type lifting" operations, may be applied to the problem of modeling apparent noncompositionality in natural language. These techniques will have to be augmented to appropriately model interactions among side effects, a subtle problem that has not been satisfactorily addressed in the programming language literature, let alone in their application to natural language. This project will relate linguistic interpretation to computational evaluation, yielding an empirically superior treatment of linguistic side effects that is more uniform and modular and thus easier to study and teach. Educational materials introducing natural-language semantics with the new framework will be developed and deployed in classroom teaching. Prototype natural-language processing systems based on the new framework will be developed to verify its predictions. By making these materials freely available, the project will contribute to the educational infrastructure, and through dissemination of papers and software, to the research infrastructure in linguistics and computer science. The project will also provide valuable research experience for undergraduates and advanced training for graduate students.