Résumé
This article presents an implemented prototype of a French-to-English translation system which goal is to ensure a syntactic quality of translation. The method employed relies on a French parser, named SYGFRAN, which in turn is based on an engine that uses Markov's algorithms applied to tree transducers. SYGFRAN produces a syntactic tree structure as output, with non terminal nodes tagged with part-of-speech constituents and dependencies. In this presentation, we show that divergence is the basis upon which we try to model the transformation path between an analyzed French sentence and its corresponding generated English sentence. We present a set of examples of divergence rules with different scopes, and a first evaluation of the prototype on two corpora: A random corpus, and a chosen corpus of,syntactically difficult' French sentences.