Coreference in a Frame Database

Abstract

This paper analyzes coreference within a frame-based semantic system. We discuss mechanisms for determining whether a new piece of knowledge (expressed as a frame) is actually a previously known referent. We propose frame database heuristics that allow new knowledge to be uniquely disambiguated for coreference so that concepts initially represented in many forms can be assimilated to those parts of the database that express the same knowledge, though perhaps in a different form. We hypothesize that these coreference mechanisms can be used by a text processing module whose control structure makes use of the organization of the text to reduce the combinatorial explosion of coreference searches in a database.