Creation of a parallel corpora from comparable corpora for the simplification of medical texts in French

The purpose of automatic simplification is to create version of texts which is easier to understand for a given targeted population. We aim at simplifying medical texts. Usually, lexicon and rules required for the simplification are acquired from parallel corpora. Since such corpora are not available for French, we propose methods for their creation from comparable corpora. Our method relies on filtering step, which purpose is to keep the best sentence candidates for alignment, and alignment step considered as categorization problem. The aim is to decide whether a pair of sentences is alignable or not. We exploit different types of features (mainly issued from lexicon and corpora) and get up to 0.97 F-measure with balanced data.