Exploiting Term Relations for Semantic Hierarchy Construction

Abstract

This paper presents a method to induce semantic taxonomies by applying the lattice theoretical technology of formal concept analysis to relations of predicates extracted from a natural language corpus. Our initial research results are in support of a future overall methodology for the semi-automatic construction of semantic hierarchies from term relations extracted from text. We describe our formal method for hierarchy construction, selection and processing of a test corpus for extracting verb-noun pairs from natural language, measurement and filtering of the resulting verb-noun term matrices for density optimization, and look at the resulting semantic hierarchies produced by FCA.