LIMSI's Contribution to the WMT'16 Biomedical Translation Task - Archive ouverte HAL Accéder directement au contenu
Communication Dans Un Congrès Année : 2016

LIMSI's Contribution to the WMT'16 Biomedical Translation Task

Résumé

The article describes LIMSI’s submission to the first WMT’16 shared biomedical translation task, focusing on the sole English-French translation direction. Our main submission is the output of a MOSES-based statistical machine translation (SMT) system, rescored with Structured OUtput Layer (SOUL) neural network models. We also present an attempt to circumvent syntactic complexity: our proposal combines the outputs of PBSMT systems trained either to translate entire source sentences or specific syntactic constructs extracted from those sentences. The approach is implemented using Confusion Network (CN) decoding. The quality of the combined output is comparable to the quality of our main system.
Fichier principal
Vignette du fichier
W16-2337.pdf (219.9 Ko) Télécharger le fichier
Origine : Fichiers éditeurs autorisés sur une archive ouverte
Loading...

Dates et versions

hal-01388658 , version 1 (03-10-2017)

Identifiants

Citer

Julia Ive, François Yvon, Aurélien Max. LIMSI's Contribution to the WMT'16 Biomedical Translation Task. First Conference on Machine Translation, Aug 2016, Berlin, Germany. pp.469-476, ⟨10.18653/v1/W16-2337⟩. ⟨hal-01388658⟩
138 Consultations
168 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More