Taxonomy Construction using Syntactic Contextual Evidence

Abstract

Taxonomies serve as the foundational structure for numerous structured, semantic knowledge resources. Recent efforts aimed at extracting taxonomic relations from text have primarily focused on collecting lexical-syntactic patterns to identify these relations through pattern matching. However, these approaches often suffer from limited coverage due to their inability to analyze context across sentences comprehensively. To address this limitation, we introduce a novel approach that leverages contextual information from terms within syntactic structures. Specifically, if the set of contexts for one term includes most of the contexts for another term, we infer a subsumption relation between the two terms. We apply this method to the task of constructing taxonomies from scratch, complementing it with an innovative graph-based algorithm for taxonomic structure induction. Our experimental results demonstrate that this approach effectively complements previous linguistic pattern matching methods and significantly enhances recall, thereby improving the overall F-measure.

Luu Anh Tuan

Assistant Professor

My research interests lie in the intersection of Artificial Intelligence and Natural Language Processing.