Nowadays, most of the statistical translation systemsare based on phrases (I.e. groups of words).In this paper we study different improvements tothe standard phrase-based translation system. Wedescribe a modified method for the phrase extractionwhich deals with larger phrases while keepinga reasonable number of phrases. We also proposeadditional features which lead to a clear improvementin the performance of the translation. Wepresent results with the EuroParl task in the directionSpanish to English and results from the evaluationof the shared task “Exploiting Parallel Textsfor Statistical Machine Translation” (ACL Workshopon Parallel Texts 2005).
展开▼