Learning to Route Languages for Multilingual Policy Optimization

Geyang Guo, Hiromi Wakaki, Yuki Mitsufuji, Alan Ritter, Wei Xu
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:38356-38378, 2026.

Abstract

Large language models (LLMs) are trained on heterogeneous multilingual corpora, yet existing policy optimization methods often implicitly restrict each training question to a single response language or rely on a fixed dominant language for supervision. We propose language-routed policy optimization (LRPO), an online reinforcement learning framework that treats language as a selectable variable. LRPO elicits multilingual rollouts for each training question and integrates their relative quality into preference-based policy updates, increasing the diversity and informativeness of training signals under the fixed rollout budget. To adaptively determine which languages to explore during reinforcement learning, we introduce a trainable language router formulated as a multi-armed bandit, balancing exploration of underutilized languages with exploitation of more informative ones. Extensive experiments show that LRPO consistently improves multilingual performance, demonstrating that adaptive language routing enables effective cross-lingual knowledge exploitation for training. We release all the resources at https://github.com/Guochry/LRPO.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-guo26v, title = {Learning to Route Languages for Multilingual Policy Optimization}, author = {Guo, Geyang and Wakaki, Hiromi and Mitsufuji, Yuki and Ritter, Alan and Xu, Wei}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {38356--38378}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/guo26v/guo26v.pdf}, url = {https://proceedings.mlr.press/v306/guo26v.html}, abstract = {Large language models (LLMs) are trained on heterogeneous multilingual corpora, yet existing policy optimization methods often implicitly restrict each training question to a single response language or rely on a fixed dominant language for supervision. We propose language-routed policy optimization (LRPO), an online reinforcement learning framework that treats language as a selectable variable. LRPO elicits multilingual rollouts for each training question and integrates their relative quality into preference-based policy updates, increasing the diversity and informativeness of training signals under the fixed rollout budget. To adaptively determine which languages to explore during reinforcement learning, we introduce a trainable language router formulated as a multi-armed bandit, balancing exploration of underutilized languages with exploitation of more informative ones. Extensive experiments show that LRPO consistently improves multilingual performance, demonstrating that adaptive language routing enables effective cross-lingual knowledge exploitation for training. We release all the resources at https://github.com/Guochry/LRPO.} }
Endnote
%0 Conference Paper %T Learning to Route Languages for Multilingual Policy Optimization %A Geyang Guo %A Hiromi Wakaki %A Yuki Mitsufuji %A Alan Ritter %A Wei Xu %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-guo26v %I PMLR %P 38356--38378 %U https://proceedings.mlr.press/v306/guo26v.html %V 306 %X Large language models (LLMs) are trained on heterogeneous multilingual corpora, yet existing policy optimization methods often implicitly restrict each training question to a single response language or rely on a fixed dominant language for supervision. We propose language-routed policy optimization (LRPO), an online reinforcement learning framework that treats language as a selectable variable. LRPO elicits multilingual rollouts for each training question and integrates their relative quality into preference-based policy updates, increasing the diversity and informativeness of training signals under the fixed rollout budget. To adaptively determine which languages to explore during reinforcement learning, we introduce a trainable language router formulated as a multi-armed bandit, balancing exploration of underutilized languages with exploitation of more informative ones. Extensive experiments show that LRPO consistently improves multilingual performance, demonstrating that adaptive language routing enables effective cross-lingual knowledge exploitation for training. We release all the resources at https://github.com/Guochry/LRPO.
APA
Guo, G., Wakaki, H., Mitsufuji, Y., Ritter, A. & Xu, W.. (2026). Learning to Route Languages for Multilingual Policy Optimization. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:38356-38378 Available from https://proceedings.mlr.press/v306/guo26v.html.

Related Material