Abstract
Word Net provides a semantic hierarchy with broad lexical coverage, which has proved sufficiently precise to boost performance at many tasks involving natural language. However, it has not yet been formalized for use in a general reasoning system. In this paper, we present such a formalization. We use a semi-automatic annotation of Word Net with lexical features - most notably the mass-count distinction - to recognize inferentially different relations between concepts. The result is a collection of 77,263 lexical-semantic axioms, which are being released for general use. We evaluate a sample of the axioms for core concepts, showing their quality to be significantly better than a baseline interpretation of Word Net.