Modèles adaptatifs pour prédire automatiquement la compétence lexicale d'un apprenant de français langue étrangère (Adaptive models for automatically predicting the lexical competence of French as a foreign language learners)

This study examines the use of supervised incremental machine learning techniques to automatically predict the lexical competence of French as a foreign language learners (FFL). The targeted learners are native speakers of Dutch having attained the A2/B1 proficiency level according to the Common European Framework of Reference for Languages (CEFR). Following recent work on lexical proficiency prediction using complexity indices, we elaborate two types of models that adapt to feedback disclosing the learners’ knowledge. In particular, we define (i) a model that predicts the lexical competence of learners having the same proficiency level and (ii) a model that predicts the lexical competence of one particular learner. The obtained models are then evaluated with respect to a baseline model, which predicts the lexical competence based on a specialised lexicon for FFL, and appear to gain significantly in accuracy (9%-17%).