second-language-acquisition

This paper proposes a multi-agent embodied conversational system that generates level-appropriate dialogue for English learners using a generate-evaluate-regenerate loop with LLMs and a CEFR classifier. A pilot study with Japanese university students showed improved level appropriateness but no statistically significant reduction in foreign language anxiety.

Dango is a 1.8B-parameter LLM trained strictly on Japanese (L1) then fine-tuned on English (L2) to study language transfer effects in second language acquisition. The model filters English contamination from the pretraining corpus and shows human-like L2 production patterns.