Exploring Natural Language Processing Methods for Finno-Ugric Langages
Résumé
This paper presents some preliminary experiments concerning the automatic processing of Finno-Ugric languages with computers. We present symbolic methods as well as machine learning ones. Given the lack of corpora for some languages , we think finite-state transducers may sometimes be the best approach where only little data are available for learning. We also consider some machine learning approaches that could be valuably applied in this context, more specifically lightly supervised techniques involving a reduced sample of annotated data and larger amounts of non annotated data. Lastly we present the LAKME project that will explore new techniques for parsing morphology-rich languages.
Origine | Fichiers produits par l'(les) auteur(s) |
---|
Loading...