Towards a More Efficient Arabic-French Translation

Even though word-to-word translation is a relatively trivial task, translation becomes more and more difficult once sentences to be translated contain compound elements such as named entities (NEs), or elements that must be kept unmodified, such as proper names. We present a system that translates Arabic NEs using local grammars. We have developed specific linguistic resources for the recognition of NEs, to which we have added a transliteration module. We then evaluate our system and show that it gives better results than other translators such as Google and Reverso. After that, we show the importance of NEs recognition in relative clause translation while comparing the obtained results with Google and Reverso. We finally argue for the viability of rule-based linguistic approaches to machine translation.