The Atayal language, one of Taiwan’s major Indigenous Austronesian languages, possesses a unique phonological and grammatical system distinct from Mandarin Chinese—the dominant medium of instruction in Taiwan’s education system. Despite the Atayal population numbering approximately 90,000, a severe shortage of qualified teachers has hindered effective instruction in the mother tongue. This study explores the development of an AI-assisted system that provides real-time evaluation and feedback on students’ oral performance in Atayal, specifically targeting pronunciation, fluency, and prosody.
Focusing on university-level beginners enrolled in an introductory Atayal course, the project addresses the pedagogical challenge of helping students produce full Atayal sentences in a short period. Oral practice was scaffolded through guided reading of Atayal texts. The AI system was trained using Wav2Vec2-Base from Hugging Face for feature extraction and Gumbel Softmax for contrastive loss prediction. Audio samples were created and labeled across six proficiency levels in three key dimensions: phoneme accuracy, fluency, and naturalness of intonation.
To evaluate system effectiveness, ten experienced Atayal language instructors reviewed the model’s performance. While the AI predictions reached approximately 70% alignment with human scoring, teachers acknowledged the system’s strong potential, especially when supported by expanded training data. Feedback highlighted its usefulness not only in assisting novice instructors but also in enhancing teachers’ own listening and language skills.
This interdisciplinary project demonstrates how AI can serve as a pedagogical ally in low-resource language education, helping to overcome teacher shortages, guide learners with individualized feedback, and support the sustainable development and intergenerational transmission of Indigenous languages.

Authors:
Yao-Rong Yun, National Tsing Hua University, Taiwan
Chingching Lu, National Tsing Hua University, Taiwan

About the Presenter(s)
Yun Yao-Rong is a third-year master's student at the Institute of Taiwan Language Research and Teaching, College of Education, National Tsing Hua University. His current project involves developing an AI-based automatic scoring and feedback system.

See this presentation on the full schedule – On Demand Schedule

Bookmark