CLASP
The Centre for Linguistic Theory and Studies in Probability

Towards Co-Inductive Models for Natural Language Semantics

Wlodek Zadrozny (UNC Charlotte and Duke U.)

In this talk we are proposing adding coinduction to the computational apparatus of semantics. This, we argue, will provide a basis for a more realistic, computationally sound, and scalable model of natural language understanding. Given that the bottom up, inductively constructed, semantic structures are brittle, and seemingly incapable of representing longer sentences or realistic dialogues, semantics is in the need of a new foundation. Coinduction, which uses top down constraints, has been successfully used in the design of operating systems and programming languages. Moreover, implicitly it has been present in text mining, machine translation, and in some attempts to model intensionality and modalities. So, there is scattered evidence it works. Since coinduction and induction can coexist, they can provide a common language and a conceptual model for research in NL understanding. We will end listing a few problems amenable to the use of coinduction, and proposing other, more challenging measures of success.