Analyse des corpus d'apprenants pour l'ELAO base sur le TAL

This paper discusses the value of using learner corpora to improve language teaching (CALL) software, and NLP tools, focussing on an error detection system to be included within a CALL software for French. We explain how errors are encoded in the corpora and present different types of analyses possible once the corpora have been annotated. We also describe error detection levels, the techniques used for automatic error detection and preliminary results. We then discuss the didactic perspective and the interface of the whole CALL software. Taking past participle agreement as an example, we also provide a concrete illustration of the interplay between learner corpora, NLP tools and didactic content. Finally, we identify key areas for further research.