Improved Statistical Machine Translation by Cross-Linguistic Projection of Named Entities Recognition and Translation

One of the existing difficulties in natural lan­ guage processing applications is the lack of appropri­ ate tools for the recognition, translation, and/or translit­ eration of named entities (NEs), specifically for lessresourced languages. In this paper, we propose a new method to automatically label multilingual parallel data for Arabic-French pair of languages with named entity tags and build lexicons of those named entities with their transliteration and/or translation in the target language. For this purpose, we bring in a third well-resourced language, English, that might serve as pivot, in order to build an Arabic-French NE Translation lexicon. Eval­ uations on the Arabic-French pair of languages using English as pivot in the transitive model showed the ef­ fectiveness of the proposed method for mining ArabicFrench named entities and their translations. Moreover, the integration of this component in statistical machine translation outperformed the baseline system.