Agenda
Lasha Abzianidze
Automating Natural Reasoning for Natural Language
In general, detecting semantic relations between the meanings of sentences requires reasoning, and it represents a substantial part of natural language understanding. In this talk, I will show how such reasoning can be modelled in a systematic way and demonstrate its implementation as a proof of concept. The underlying methodology is inspired by the studies in formal logic and semantics. The resulting automated reasoning system, unlike ordinary automated theorem provers, is more “natural” as it is specifically designed for reasoning with natural language. I will also introduce a task of natural language inference (i.e. a benchmark that evaluates natural language processing systems on reasoning with natural language text), and present the results of the system on it. (A spoiler: an online demo of the reasoning system can be found at bit.ly/LanPro)