Large Reasoning Models (LRMs) have achieved remarkable performance on complex reasoning tasks by adopting the "think-then-answer" paradigm, which enhances both accuracy and interpretability. However, current LRMs exhibit two critical limitations when processing non-English languages: (1) They often struggle to maintain input-output language consistency; (2) They generally perform poorly with wrong reasoning paths and lower answer accuracy compared to English. These limitations significantly degrade the user experience for non-English speakers and hinder the global deployment of LRMs. To address these limitations, we propose M-Thinker, which is trained by the GRPO algorithm that involves a Language Consistency (LC) reward and a novel Cross-lingual Thinking Alignment (CTA) reward. Specifically, the LC reward defines a strict constraint on the language consistency between the input, thought, and answer. Besides, the CTA reward compares the model's non-English reasoning paths with its English reasoning path to transfer its own reasoning capability from English to non-English languages. Through an iterative RL procedure, our M-Thinker-1.5B/7B models not only achieve nearly 100% language consistency and superior performance on two multilingual benchmarks (MMATH and PolyMath), but also exhibit excellent generalization on out-of-domain languages.
If you find this work useful, please consider citing our paper:
@misc{zhang2025thinknativelyunlockingmultilingual,
title={Think Natively: Unlocking Multilingual Reasoning with Consistency-Enhanced Reinforcement Learning},
author={Xue Zhang and Yunlong Liang and Fandong Meng and Songming Zhang and Kaiyu Huang and Yufeng Chen and Jinan Xu and Jie Zhou},
year={2025},
eprint={2510.07300},
archivePrefix={arXiv},
primaryClass={cs.CL},
url={https://arxiv.org/abs/2510.07300},
}
Training hyperparameters
The following hyperparameters were used during training:
learning_rate: 5e-07
train_batch_size: 1
eval_batch_size: 8
seed: 42
distributed_type: multi-GPU
num_devices: 8
gradient_accumulation_steps: 32
total_train_batch_size: 256
total_eval_batch_size: 64
optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
