AI-enhanced graded viewers: Creating level-appropriate video content for language learning

Just as graded readers provide language learners with level-appropriate texts to develop reading skills, 'graded viewers' can deliver customised audiovisual content for targeted listening practice. The widespread use of language learner focused video content however is often hindered by a limited selection and a lack of engaging, level-appropriate content, due to its significant cost of production, limiting effective self-study. To address this gap, the author has developed a proof-of-concept AI-powered ‘graded viewer’ system that generates short customisable television-style videos and auto-graded comprehension quizzes. The system integrates Large Language Models (LLMs) for script generation, real-time 3D animation for visuals, and neural voice synthesis for dialogue, allowing users to create content tailored to their CEFR level across five different genres. Preliminary feedback from 20 student users indicates that the system successfully generates comprehensible input and its level of customisation is highly engaging. However, challenges in maintaining narrative coherence and ensuring consistent quality control remain. This on-going project seeks to demonstrate the potential for AI-generated graded viewers as a new category of autonomous, user-driven language learning resources that have the potential to bridge the gap between engaging authentic media and pedagogically sound educational materials.