Logical Story Representations via FrameNet + Semantic Parsing

Abstract

We present a means of obtaining rich semantic representations of stories by combining neural FrameNet identification, a formal logic-based semantic parser, and a hierarchical event schema representation. The final schematic representation of the story abstracts constants to variables, preserving their types and relationships to other individuals in the story. All identified FrameNet frames are incorporated as temporally bound ``episodes'' and related to one another in time. The semantic role information from the frames is also incorporated into the final schema's type constraints. We describe this system as well as its possible applications to question answering and open-domain event schema learning.

Tasks

Reproductions