Traitement automatique des entités nommées en arabe : détection et traduction

The recognition of Arabic Named Entities (NE) is a potentially useful preprocessing step for many Natural Language Processing Applications, such as Statistical Machine Translation (SMT). Due to peculiarities of the written Arabic language, this task is however rather challenging. In this paper, we present a detailed study of Arabic NEs in the context of a SMT system. We present our statistical NE recognition system (NERAr), and its various evolutions. NERAr was then used as a processing step, thus enabling us to incorporate external linguistic knowledge into the SMT system. Several strategies for performing this integration are explored. Automatic evaluations, corroborated by manual inspections, indicate a small improvement of the translation quality of NEs, and a reduction of the errors caused by out-of-vocabulary words. MOTS-CLÉS : entités nommées, traduction automatique, traitement automatique de l’arabe.