Utilisation des réseaux de neurones récurrents pour la projection interlingue d'étiquettes morpho-syntaxiques à partir d'un corpus parallèle
Résumé
In this paper, we propose a method to automatically induce linguistic analysis tools for languages that have no labeled training data. This method is based on cross-language projection of linguistic annotations from parallel corpora. Our method does not assume any knowledge about foreign languages, making it applicable to a wide range of resource-poor languages. No word alignment information is needed in our approach. We use Recurrent Neural Networks (RNNs) as cross-lingual analysis tool. To illustrate the potential of our approach, we firstly investigate Part-Of-Speech (POS) tagging. Combined with a simple projection method (using word alignment information), it achieves performance comparable to the one of recently published approaches for cross-lingual projection. Mots-clés : Multilinguisme, transfert crosslingue, étiquetage morpho-syntaxique, réseaux de neurones récurrents.
Origine | Fichiers éditeurs autorisés sur une archive ouverte |
---|
Loading...