Sub-Symbolic Mapping of Cyc Microtheories in Data-Driven 'Conceptual' Spaces

Abstract

The presented work aims to combine statistical and cognitive-oriented approaches with symbolic ones so that a conceptual similarity relationship layer can be added to a Cyc KB microtheory. Given a specific microtheory, a LSA-inspired conceptual space is inferred from a corpus of texts created using both ad hoc extracted pages from the Wikipedia repository and the built-in comments about the concepts of the specific Cyc microtheory. Each concept is projected in the conceptual space and the desired layer of sub-symbolic relationships between concepts is created. This procedure can help a user in finding the concepts that are “sub-symbolically conceptually related” to a new concept that he wants to insert in the microtheory. Experimental results involving two Cyc microtheories are also reported.