Building a concept hierarchy from corpus analysis

Abstract

Corpus analysis is today at the heart of building Terminological Knowledge Bases (TKBs). Important terms are usually first extracted from a corpus and then related to one another via semantic relations. This research brings the discovery of semantic relations to the forefront to allow the discovery of less stable lexical units or unlabeled concepts, which are important to include in a TKB to facilitate knowledge organization. We suggest a concept hierarchy made of concept nodes defined via a representational structure emphasizing both labeling and conceptual representation. The Conceptual Graph formalism chosen for conceptual representation allows a compositional view of concepts, which is relevant for their comparison and their organization in a concept lattice. Examples manually extracted from a scuba-diving corpus are presented to explore the possibilities of this approach. Subsequently, steps toward a semi-automatic construction of a concept hierarchy from corpus analysis are presented to evaluate their underlying hypothesis and feasibility.