- Casas-Tost, Helena|||0000-0002-4023-9070
- González Torrents, Isabel|||0009-0007-8495-8396
- Rovira-Esteva, Sara|||0000-0001-7647-6417
- Vargas-Urpi, Mireia|||0000-0001-6302-581X
This presentation investigates the efficacy of Generative AI (GenAI) in developing reading comprehension exercises for beginner-level learners of Chinese. We conducted a comparative analysis of two prominent models, ChatGPT-4o and DeepSeek-R1, employing a novel assessment tool designed to evaluate the linguistic quality, pedagogical relevance, and cost-effectiveness of both the AI-generated texts and their accompanying questions. Our findings indicate that while both systems produced outputs of broadly similar quality, native speaker evaluations gave a higher overall rating to texts from ChatGPT-4o. Crucially, both AIs demonstrated significant utility for rapidly generating base texts, offering substantial time savings with minimal required revision. However, a key limitation was identified in the systems' ability to formulate pedagogically sound reading comprehension questions, a task that still necessitates human expertise. The developed assessment tool proved highly effective in helping educators quickly identify which AI-generated materials were classroom-ready, which needed revision, and which should be rejected. In conclusion, this study underscores the dual reality of GenAI in language instruction: it is a powerful tool for efficiency, but its effective application remains dependent on professional oversight to ensure pedagogical quality.