Crosslingual Reasoning through Test-Time Scaling

ICLR 2026 Conference Submission13306 Authors

Keywords: LLM reasoning, crosslingual transfer

Abstract:

Reasoning capabilities of large language models are primarily studied for English, even when pretrained models are multilingual.
In this work, we investigate to what extent English reasoning finetuning can generalize across languages.
First, we find that sequential test-time scaling for English-centric reasoning language models (RLMs) through longer chain-of-thoughts (CoTs) improves multilingual mathematical reasoning across many languages including low-resource languages, to an extent where they outperform models *twice their size*.
Second, we reveal that while English-centric RLM's CoTs are naturally predominantly English, they consistently follow a *quote-and-think* pattern to reason about quoted non-English inputs.
Third, we discover an effective strategy to control the language of long CoT reasoning, and we observe that models reason better and more efficiently in high-resource languages.
Overall, we demonstrate the potentials, study the mechanisms, and outline the limitations of crosslingual generalization of English reasoning test-time scaling. We conclude that practitioners should let English-centric RLMs reason in high-resource languages, while further work is needed to improve reasoning in low-resource languages.

Primary Area: foundation or frontier models, including LLMs

Submission Number: 13306