Code-switched training improves domain reasoning in target languages

From Unverifiable Domain Text to Cross-Lingual RLVR via Code-Switched Task Synthesis

Computation and Language

Summary

Understanding specialized information like medical or cybersecurity knowledge can be hard for AI models in languages other than English, due to lack of good training tasks. The authors introduce a method that creates language-learning tasks mixing English and a target language by transforming English texts into verifiable questions and answers. Training models with these mixed-language tasks helps them better understand and reason in the target language while keeping good performance in English. This approach was tested on Korean, medical, and cybersecurity text across multiple languages, showing improved results compared to just using single or bilingual data.

What this means in practice

  • For multilingual ai developers: Create language training tasks mixing English and target languages for better domain reasoning without requiring new target-language labeled data.
  • For language technology companies: Enhance multilingual chatbot or AI assistant performance in specialized fields like medicine or security by improving reasoning in underrepresented languages.$Commercial implications: Enables selling advanced multilingual domain-specific AI services that surpass current monolingual or bilingual approaches.

Authors

Cheonbok Park, Jihwan Kim, Shinyoung Joo, Joosung Lee, Hwiyeol Jo, Jeonghoon Kim, Jaegul Choo

Abstract

Reinforcement Learning with Verifiable Rewards (RLVR) can improve reasoning in large language models, but specialized non-English domains expose a data mismatch. Raw English domain text is often available, while target-language RLVR tasks with verifiable rewards are scarce. We introduce code-switched task synthesis for cross-lingual RLVR, a data-centric approach that derives target-language and code-switched RLVR task variants from otherwise unverifiable English domain text. Using fill-in-the-middle multiple-choice QA as a seed task format, we synthesize automatically verifiable tasks from domain passages and reshape them into cross-lingual variants. These variants preserve the same evidence-answer relation but change how English and target-language evidence are presented during RLVR. On Korean medical benchmarks and cybersecurity benchmarks across four target languages, code-switched training outperforms matched monolingual and bilingual-mixture baselines on average target-language performance while maintaining comparable English performance. Additional analyses provide evidence of better target-language input understanding and stronger cross-lingual alignment in both model outputs and internal representations. These results suggest that code-switched training helps target-language inputs access domain reasoning learned more strongly in English.
View PDFOpen arXiv