LANG: Reinforcement Learning for Multilingual Reasoning with Language-Adaptive Hint Guidance
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866916036784160768 |
|---|---|
| author | Fan, Yuchun Li, Bei Li, Peiguang Wang, Yilin Mu, Yongyu Yang, Jian Chen, Xin Weng, Rongxiang Wang, Jingang Cai, Xunliang Zhu, Jingbo Xiao, Tong |
| author_facet | Fan, Yuchun Li, Bei Li, Peiguang Wang, Yilin Mu, Yongyu Yang, Jian Chen, Xin Weng, Rongxiang Wang, Jingang Cai, Xunliang Zhu, Jingbo Xiao, Tong |
| contents | Reinforcement learning has proven effective for enhancing multi-step reasoning in large language models (LLMs), yet its benefits have not fully translated to multilingual contexts. Existing methods struggle with a fundamental trade-off: prioritizing input-language consistency severely hampers reasoning quality, while prioritizing reasoning often leads to unintended language drift toward English. We address this challenge with LANG, a novel framework that leverages language-conditioned hints to guide exploration in non-English reasoning tasks. Our method incorporates two key mechanisms to prevent dependency on these hints: a progressive decay schedule that gradually withdraws scaffolding, and a language-adaptive switch that tailors learning horizons to specific language difficulties. Empirical results on challenging multilingual mathematical benchmarks reveal that LANG substantially enhances reasoning performance without compromising language consistency. Moreover, we show that our framework generalizes beyond mathematics, fostering more consistent language alignment across model layers |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_22567 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | LANG: Reinforcement Learning for Multilingual Reasoning with Language-Adaptive Hint Guidance Fan, Yuchun Li, Bei Li, Peiguang Wang, Yilin Mu, Yongyu Yang, Jian Chen, Xin Weng, Rongxiang Wang, Jingang Cai, Xunliang Zhu, Jingbo Xiao, Tong Computation and Language Reinforcement learning has proven effective for enhancing multi-step reasoning in large language models (LLMs), yet its benefits have not fully translated to multilingual contexts. Existing methods struggle with a fundamental trade-off: prioritizing input-language consistency severely hampers reasoning quality, while prioritizing reasoning often leads to unintended language drift toward English. We address this challenge with LANG, a novel framework that leverages language-conditioned hints to guide exploration in non-English reasoning tasks. Our method incorporates two key mechanisms to prevent dependency on these hints: a progressive decay schedule that gradually withdraws scaffolding, and a language-adaptive switch that tailors learning horizons to specific language difficulties. Empirical results on challenging multilingual mathematical benchmarks reveal that LANG substantially enhances reasoning performance without compromising language consistency. Moreover, we show that our framework generalizes beyond mathematics, fostering more consistent language alignment across model layers |
| title | LANG: Reinforcement Learning for Multilingual Reasoning with Language-Adaptive Hint Guidance |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2605.22567 |