Filtering parallel texts to improve translation model and cross-language information retrieval

Jian Cai


Le 2 mai 2001 à 11 h 30

Salle 3195, Pavillon André-Aisenstadt

One approach to cross-language information retrieval (CLIR) is to use a statistical translation model trained on parallel texts. In a previous project, several sets of parallel texts have been mined automatically from the Web. However, all the texts are not truly parallel. In order to increase the quality of the translation model, we tested several filtering approaches to parallel texts during and after sentence alignment. These approaches use either a bilingual dictionary, sentence lengths, the proportion of non-aligned sentences, or a combination of them. The experimental results on Chinese-English parallel texts show that after the filtering, the quality of the translation model, as well as CLIR effectiveness, are increased significantly.

Pour recevoir les annonces hebdomadaires par courriel, visitez

Liste des autres séminaires