Explainable Personalised Automated Writing Feedback for L2 Dutch

Effective writing skills are essential to succeed in professional and academic settings. In Flanders, highly-educated non-native individuals are entering the competitive labour market while struggling with writing in Dutch. From research we know that detailed, constructive feedback fosters writing development. However, providing feedback is a time-consuming task, and therefore automated writing support is considered a solution. Currently, this support is outsourced to generative AI, but chatbots mainly (re)write texts and are not designed to provide pedagogically sound feedback. Moreover, studies regarding their impact on the learning and writing process are lacking and their bias to English and black-box nature demands more research.

EXPLAINED will develop a pedagogically sound writing support system for L2 Dutch learners. State-of-the-art open-source large language models will be used to detect writing problems at both sentence and text level. By exploiting generative AI, error detection will be directly complemented with qualitative and interpretable feedback, tailored to each learner’s native language and Dutch proficiency level. The validity of the proposed model will be tested by both investigating its portability to other European languages and by conducting use case experiments.

This system will motivate L2 learners to correct their own errors and hone their writing skills, empowering them in academic studies or a professional setting.