Skip to Main content Skip to Navigation
Master Thesis

Constitution et évaluation d'un jeu de données linguistiques en français pour l'analyse des fonctions lexicales encodées dans les modèles neuronaux de type FlauBERT

Abstract : Each language is made up of its own words. In most cases, these are polysemic, they have several meanings. Modeling polysemy in Automatic Language Processing is a difficult task when it comes to word vectors and traditional word embeddings systems have some difficulties in dealing with polysemy. Using FlauBERT, which is a new language model developed in 2019, we will see that it is now easier to deal with polysemy, especially with contextualized word vectors. The entire context of a sentence is taken into account by FlauBERT in order to represent each word as a vector. After a brief analysis of the different domains involved, I will present in this paper the different experiments I have performed using FlauBERT word vectors.
Document type :
Master Thesis
Complete list of metadatas

Cited literature [38 references]  Display  Hide  Download

https://dumas.ccsd.cnrs.fr/dumas-02978401
Contributor : Uga - Bulles <>
Submitted on : Monday, October 26, 2020 - 2:01:54 PM
Last modification on : Tuesday, November 24, 2020 - 4:00:19 PM

File

BELLUE Vincent_M2_IDL.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : dumas-02978401, version 1

Citation

Vincent Bellue. Constitution et évaluation d'un jeu de données linguistiques en français pour l'analyse des fonctions lexicales encodées dans les modèles neuronaux de type FlauBERT. Sciences de l'Homme et Société. 2020. ⟨dumas-02978401⟩

Share

Metrics

Record views

14

Files downloads

16