What Does Post-Training Change in Multilingual Reasoning?

2026-09-29

AI Summary

This study addresses the significant performance gap of open-source reasoning models in multilingual scenarios, where non-English accuracy lags substantially behind English. We systematically audit the multilingual reasoning bottlenecks of Qwen3, revealing that post-training constitutes the dominant limiting factor. To overcome this, we propose a reinforcement learning framework incorporating multilingual supervised fine-tuning and language-term rewards, which effectively mitigates non-termination loops and language fallback issues. This work establishes a constructive post-training pathway that transfers English-centric capabilities to multilingual reasoning, significantly improving both answer completeness and accuracy in non-English settings.

Abstract

Open-source reasoning models provide unequal access to reasoning capability across languages. When a model can solve a problem but cannot deliver a complete solution in the user's language, language becomes an access barrier rather than merely a source of performance variation. We audit Qwen3 checkpoints on competition-mathematics tasks in eleven languages. Across the ten non-English languages, only 15.4-17.9% of problems receive a correct, terminating solution with visible reasoning in the requested language in any of 16 samples, compared with 92.9% in English. To identify the source of this disparity, we evaluate thirteen endpoints from one model family, spanning released checkpoints, multilingual supervised fine-tuning (SFT) at two scales, controlled SFT ablations, and three reinforcement-learning (RL) reward formulations. We jointly track correctness, language adherence, termination, and delivery efficiency. The dominant bottleneck shifts across post-training stages. Released models often reason in English. Multilingual SFT restores target-language reasoning, but accuracy declines across multilingual, English-only, and single-language SFT runs, showing that this cost is not specific to multilingual mixing; non-English reasoning traces additionally become prone to non-terminating loops. RL restores termination in both arms at no cost in accuracy, but only the arm whose reward includes a language term delivers: rewarding correctness alone returns the model to English. Together, these stages establish a constructive post-training path from English-pivoted capability to multilingual reasoning that is reliably delivered.

Problem

• multilingual reasoning
• post-training
• language disparity
• open-source reasoning models
• reinforcement learning

Innovation

• Multilingual Reasoning
• Post-Training
• Supervised Fine-Tuning
• Reinforcement Learning
• Reward Formulation