Compression de vocabulaire de sens grâce aux relations sémantiques pour la désambiguïsation lexicale

Abstract : In Word Sense Disambiguation (WSD), supervised approaches are predominant in evaluation cam- paigns. The limited quantity of such corpora however restricts the coverage and the performance of these systems. In this article, we present two new methods that tackle this problem by exploiting the semantic relationships between senses such as synonymy, hypernymy and hyponymy, in order to compress the sense vocabulary of WordNet, and thus reduce the number of different sense tags that must be observed to disambiguate all words of the lexical database. Our methods greatly reduce the size of neural WSD models, with the benefit of improving their coverage without additional training data, and without impacting their precision. In addition to our methods, we present a neural WSD system which relies on the recent advances in contextualized word embeddings in order to achieve results that significantly outperform the state of the art on all WSD evaluation tasks.
Complete list of metadatas

Cited literature [33 references]  Display  Hide  Download

https://hal.archives-ouvertes.fr/hal-02176195
Contributor : Didier Schwab <>
Submitted on : Sunday, July 7, 2019 - 7:52:59 PM
Last modification on : Thursday, July 11, 2019 - 1:06:26 AM

File

TALN_2019.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-02176195, version 1

Collections

LIG | UGA

Citation

Loïc Vial, Benjamin Lecouteux, Didier Schwab. Compression de vocabulaire de sens grâce aux relations sémantiques pour la désambiguïsation lexicale. TALN 2019 (Conférence sur le Traitement Automatique des Langues Naturelles), Jul 2019, Toulouse, France. ⟨hal-02176195⟩

Share

Metrics

Record views

18

Files downloads

23