ETRI Unveils AI Korean Teacher with Advanced Learning Capabilities

October 28, 2026
ETRI Unveils AI Korean Teacher with Advanced Learning Capabilities
  • ETRI in South Korea has developed a Korean AI teacher that teaches Korean to foreign learners by combining free conversation with structured progress tracking.

  • ETRI also announces a Korean-language teaching AI capable of engaging learners in free-flow conversations while ensuring all lesson content is covered.

  • The system uses three mechanisms: a stage-detection component to decide when to advance, a division of labor where familiar expressions are handled by a smaller model (3 to 13 billion parameters) and unfamiliar terms are passed to general-purpose AIs like GPT, and a filter to prevent repeating questions.

  • In evaluations, the AI achieved the lowest conversational error rate (18.6%) compared with GPT-4o-mini (25.4%) and GPT-5.1 (23.7%), while learners preferred ETRI’s system in eight of ten scenarios, though GPT models performed better at avoiding repetition and varying expressions.

  • Current capabilities include situational speaking exercises, free conversation, reading-comprehension-based dialogue, and automated writing assessment, with the pilot expanding toward a comprehensive reading, writing, and speaking tutor.

  • Future plans include refining performance through user data collection, optimizing model role allocation, expanding to reading and writing, and offering a personalized AI Korean teacher tailored to individual proficiency and study history.

  • The glasses system connects ETRI’s small language model with GPT technology, alternating responses between the two models.

  • Lim Su-jong, head of ETRI's Language Intelligence Research Laboratory, said the results show competitiveness through external blind evaluation and highlighted ongoing development toward trusted AI education technology for Korean learners worldwide.

  • ETRI aims to evolve the technology into a personalized education service tailored to each learner’s proficiency and study history for global use by Korean learners.

  • Leaders emphasize solving fluency gaps by providing an AI native-speaker-like partner and expanding the technology into a global AI education tool for Korean learners.

  • The system blends ETRI’s in-house small language models (3B–13B) with general-purpose AI like GPT, routing unfamiliar terms to external models while handling familiar expressions in-house to improve consistency and reduce data sent overseas.

  • A novel topic-guided free-conversation approach keeps learners on learning objectives while allowing natural dialogue, with the AI controlling lesson steps without making them explicit to the learner.

Summary based on 2 sources


Get a daily email with more AI stories

More Stories