This study belongs to the field of automatic scoring of texts written in a foreign language. It aims to evaluate in depth the utility to take into account automatic measures of the phraseological competence. The phraseological features were obtained by assigning to each bigram in a text several association scores computed on the basis of a large native corpus. The analyzes performed on three datasets indicate that the phraseological features are useful for estimating text quality and that they are complementary to simpler lexical measures. External validation, carried out by learning and testing the predictive model on very different sets of texts, shows that these indices have a high degree of generalizability.
Bestgen, Y. (2016). Validation interne et externe d’indices phraséologiques pour l’évaluation automatique de textes rédigés en anglais langue étrangère. Traitement automatique des langues, 57, 91-115. https://hdl.handle.net/2078.5/174601 (Original work published 2016)