Commonsense Knowledge Acquisition Using Compositional Relational Semantics

Abstract

A method for the acquisition of commonsense knowledge based on instantiations of metarules is presented. The metarules refer to some properties and objects that have those properties. Metarules are instantiated by automatically identifying objects that have those properties. In order to increase the applicability of a commonsense property to objects, composition of semantic relations is used. The method has been implemented and tested over WordNet. Results show that a commonsense metarule can produce many knowledge base axioms.