A comparative study of hypernymic patterns for knowledge extraction.

Abstract

Emerging semi-automatic knowledge extraction tools make it possible for terminologists to efficiently locate contexts in text that illustrate the meaning and usage of terms. Semi-automatic knowledge extraction makes use of a pre-programmed set of linguistic structures (patterns) which frequently show relationships between concepts. This thesis contributes to the linguistic research on patterns that is needed for tool development. More specifically, this thesis focuses on the patterns for one specific relation: hyperonymy. The research involved two comparative studies of hypernymic patterns, one intra-linguistic (French) and the other inter-linguistic (French-English). The first study involved discovering French patterns in two separate domains, and comparing their nature and frequency between domains. The second study involved comparing the stylistic differences between the French patterns discovered and a pre-existing set of English patterns. This research has revealed that comparative linguistic research on hypernymic patterns can provide useful information for developing terminological knowledge extraction tools.