Amélioration de l'alignement et de la traduction statistique par utilisation de corpus parallèles multilingues
Camelia Ignat · Serveur des thèses de l'Université de Strasbourg (University of Strasbourg) · 2009
This thesis presents the compilation of a highly multilingual parallel corpora (JRC-Acquis) and its usage to improve statistical alignment and translation by triangulation, which is the process of translating from a source to a target language via an intermediate third language (pivot). We explore heuristics to improve alignment and translation using multilingual, parallel, sentence-aligned corpora in several pivot languages. Our study offers two main methods utilizing a pivot language to create a translation model, with a procedure for combining translation systems for multiple pivot languages. We present experiments showing that multilingual, parallel text in twenty-two languages can be used in this framework to improve statistical translation.