Definition extraction using sense-based embeddings

Abstract

De nition Extraction is the task to identify snippets of free text in which/na term is de ned. While lexicographic studies have proposed di erent de nition typologies and categories, most NLP tasks aimed at revealing word or concept meanings have traditionally dealt with lexicographic (encyclopedic) de nitions, for example, as a prior step to ontology learning or automatic glossary construction. In this paper we describe and evaluate a system for De nition Extraction trained with features/nderived from two sources: Entity Linking as provided by Babelfy, and semantic similarity scores derived from sense-based embeddings. We show that these features have a positive impact in this task, and report state-of-the-art results over a manually validated benchmarking dataset.