Linking FrameNet to the SUMO Ontology
Jan Scheffczyk, Adam Pease and Michael Ellsworth
International Conference on Formal Ontology in Information Systems (FOIS 2006)
Baltimore, Maryland (USA), November 9-11, 2006
Abstract
Deductive reasoning with natural language requires combining semantically rich lexical resources with world knowledge provided by ontologies and databases. While great progress has been made in natural-language retrieval tasks, using natural language to support deep computer-based reasoning has progressed more slowly. Major obstacles have been the lack of large lexicons, large formal ontologies and linguistic frames, and most importantly, the relationships among these products. The Suggested Upper Merged Ontology (SUMO) is a large formal ontology coded in first order logic. FrameNet is a large lexical resource based on frame semantics, which encodes language as interrelated semantic Frames (types of predication) with frame element (FE) arguments. In this paper we describe how we are creating bindings from FrameNet to SUMO. In contrast to other lexicon-ontology connections, we express our bindings in KIF, the language of SUMO. This supports the use of arbitrary class expressions and axioms, which are necessary given inherent differences between ontologies and lexical resources. As a first step we have linked the FrameNet semantic types (ST) to SUMO. Our objective is to link an ST to SUMO classes equivalent to the ST, without restricting the links to any particular domain. Based on these links, we have developed a semi-automatic approach to link FrameNet frame elements (FE) to SUMO classes, taking advantage of pre-existing mappings from the WordNet lexicon to SUMO. This allows us to develop domain-specific links from FEs to SUMO by examining annotated examples from a particular domain (e.g., Weapons of Mass Destruction). We thus provide restricted, ontology-based types on the fillers of FEs, which should help semantic parsers both with word sense disambiguation of predicators and identifying which pieces of a sentence fill FEs. By relating FrameNet and SUMO we have realized significant benefits, such as identifying areas which can be improved in both products.