Les textes générés par l'IA comme modèle pédagogique en classe de FLE:

Aspects théoriques et pratiques

Résumé

New generative AI (GenAI) tools, machine translation and automatic text production (ATP) tools like ChatGPT are being used more and more in language classrooms, by teachers and learners alike. Integrating the output of these systems as a model for students raises several questions, some of which we discuss and address in this article. We think the way that data-driven learning was successfully introduced in language learning and teaching (LLT) is an interesting basis to consider the introduction of GenAI tools in the classroom. The first part of this text discusses theoretical principles linked to the use of these technologies by language learners. The second part presents an action-research study in which we compared the use of an ATP tool (Copilot) to the use of a corpus (Corpus Leipzig - French) in a B1 French as a foreign language vocabulary class. Students were divided in two groups and each group used one tool to find collocations for a set of words. The results show no significant difference in the way participants evaluated the usefulness of the tool and we found that the two groups gathered a similar number of correct items. However, the level of difficulty of the words selected by participants differed between the two groups: Copilot provided more advanced vocabulary (B2 and above) and students choosing from the corpus tended to pick words below or at their level. In conclusion, it seems that for this task ATP tools can successfully be used but do not offer any real added value to other tools that have been widely available for some time, namely corpora.