Annotation automatique d'erreurs dans des textes d'apprenants du français (FLE) : développement d'une ressource annotée en erreurs (approches linguistiques et techniques)

Details

Supervisors

François, Thomas ; Souza Wilkens, Rodrigo

Faculty

Faculté de philosophie, arts et lettres

Degree label

Master [120] en linguistique, à finalité spécialisée: traitement automatique du langage

Abstract

enThe study delves into the French orthography through a two-fold approach: the creation of annotated learner texts and the utilization of artificial intelligence (AI) models for error detection. By bridging the realms of linguistics and technology, this research seeks to establish comprehensive resources that facilitate both linguistic analysis and pedagogical advancement. The study unfolds with the implementation of successful annotation campaigns, yielding meticulously annotated French learner texts. These texts serve for training AI models (Transformers' architecture : BERT) designed to detect and categorize errors. Notably, the study reflects on the encouraging performance of these AI models, their potential in reshaping pedagogical paradigms, and their alignment with the intricacies of the French language. Furthermore, the study offers valuable insights into the refinement of typologies for error classification and the adaptation of AI models to accommodate the unique challenges posed by French orthography.