Abstract:

Object clitic pronouns (especially third person ones) in French are problematic for second and foreign language learners. Several researchers, such as Wust (2009), have thus uncovered the fact that French L2 learners often resort to avoidance strategies that allow them to not use these forms even if this use allows them to lighten the discourse (written or oral) by avoiding repetition. This is one of the reasons why we were interested in technological tools that could help these learners apprehend these clitics. We therefore conducted a study with a tripartite goal: to uncover a corpus of L2 French productions focusing on clitics, to use this corpus to train a state-of-the-art deep learning model (CamemBERT), and to implement the trained model to detect learners' errors when they produce the forms under study. In evaluating this model, it was found to be over 99% reliable. Moreover, by testing it on sentences with different turns of phrase from those encountered during its training, the model manages to detect errors with the same degree of reliability. This model constitutes a considerable advance in the automatic processing of interlanguage and can be used to develop tools for learners of L2 French.

Keywords:

French L2, object clitics, deep learning, CamemBERT.