French to Arabic machine translation: the specificity of language couples

French to Arabic Machine Translation: the Specificity of Language Couples Haytham Alsharaf 1, Sylviane Cardey2 and Peter Greenfield3 1French Language Unit, 2,3Centre Tesnière 1Kuwait University, 2,3Université de Franche-Comté, France 1alsharaf2001@yahoo.com, 2sylviane.cardey@univ-fcomte.fr, 3peter.greenfield@univ-fcomte.fr Abstract. We present an approach to the machine translation of French to Arabic in which we explain each of the different steps and some of the problems encountered at each level of linguistic analysis together with their solutions. We show that for high quality translation, an approach which is both global and proper to the couple French to Arabic is necessary, and we compare our approach with existing ones (direct, transfer, pivot, statistical). The fact that we have two languages which are linguistically distant requires that certain linguistic phenomena that are specific to the couple must be analysed, such phenomena not necessarily being found in other language couples. 1. Introduction Approaches to machine translation from French to Arabic are rare. Certainly, translation memory systems do exist (as for example the commercial system An-Nakel Al-Arabi from French to Arabic from CIMOS (Paris)1) but the limitations of such systems are now well known to all specialists working in the domain. In this paper we present an approach which we have implemented specially for the machine translation of specialty languages from French to Arabic. At the outset of the project, existing approaches (see for example [1], [2]) were examined (direct, transfer, pivot, statistical) but it became apparent that each presented limits and it became increasingly evident that it was necessary to devise an approach specific to this particular language couple. The fact that we have here two languages that are very distant linguistically means that, from the linguistic point of view, certain linguistic operations which are specific to this couple will not necessarily be found in other couples, whether or not one or other of French or Arabic is present as either source or target. The approach that we describe is thus unique for the couple French to Arabic; and it follows that it is not valid for the inverse couple Arabic to French. Reversing the 1 http://www.cimos.com/index.asp?src=fiche "Highlights: ... • Alignement de mémoire et de phrase de traduction" translation steps will not result in a machine translation system of the couple Arabic to French because to have such a system would necessitate certain linguistic operations that are not present for French to Arabic. Indeed, this consideration of direction in translation explains our use of the term "language couple" rather than "language pair", by analogy with "couple" in statics in which direction is inherent. Other work that we have carried out vindicates our findings concerning the uniqueness of couples for linguistically distant language machine translation (for example Chinese to French [3], Korean to French [4], both Chinese to French and French to Arabic [5]). It is our opinion that if machine translation is to succeed, and by success we mean quality machine translation, it is necessary to limit one's ambitions by dealing with one couple of languages at a time, this …

Content truncated to 500 words