Extending WordNet with UFO foundational ontology

Abstract

WordNet is a large lexical database used by an uncountable number of applications for computational linguistics. Many proposals have attempted to better describe it in a semantic perspective, especially addressing synonymy, taxonomy and mereology properties, which led to very good results in domain-specific applications. A philosophical shift on this semantic description could, however, improve the scope of these results across different domains. In this direction, this work extends WordNet’s semantic knowledge by addressing philosophical meta-properties. Specifically, we apply the notion of Semantic Types to propose mapping rules between the noun synsets of Wordnet and the top-level constructs of a foundational ontology. For this task we have chosen the Unified Foundational Ontology (UFO), which explicitly exposes philosophical meta-properties of concepts in its structure, leading to a well-founded semantically-enriched version of Wordnet. The proposed rules were validated through an experiment over approximately 5,200 sample mappings, obtaining an average accuracy of 93.5% Furthermore, to show its applicability, the proposal was applied to the task of automatically learning a well-founded domain ontology.