Saved in:
| Main Authors: | , , , , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2025
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2505.05408 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866909604903911424 |
|---|---|
| author | Yong, Zheng-Xin Adilazuarda, M. Farid Mansurov, Jonibek Zhang, Ruochen Muennighoff, Niklas Eickhoff, Carsten Winata, Genta Indra Kreutzer, Julia Bach, Stephen H. Aji, Alham Fikri |
| author_facet | Yong, Zheng-Xin Adilazuarda, M. Farid Mansurov, Jonibek Zhang, Ruochen Muennighoff, Niklas Eickhoff, Carsten Winata, Genta Indra Kreutzer, Julia Bach, Stephen H. Aji, Alham Fikri |
| contents | 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 with long chain-of-thoughts (CoTs) can generalize across languages. First, we find that scaling up inference compute for English-centric reasoning language models (RLMs) 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. Finally, we observe poor out-of-domain reasoning generalization, in particular from STEM to cultural commonsense knowledge, even for English. 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 and out-of-domain contexts. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2505_05408 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Crosslingual Reasoning through Test-Time Scaling Yong, Zheng-Xin Adilazuarda, M. Farid Mansurov, Jonibek Zhang, Ruochen Muennighoff, Niklas Eickhoff, Carsten Winata, Genta Indra Kreutzer, Julia Bach, Stephen H. Aji, Alham Fikri Computation and Language Artificial Intelligence Machine Learning 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 with long chain-of-thoughts (CoTs) can generalize across languages. First, we find that scaling up inference compute for English-centric reasoning language models (RLMs) 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. Finally, we observe poor out-of-domain reasoning generalization, in particular from STEM to cultural commonsense knowledge, even for English. 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 and out-of-domain contexts. |
| title | Crosslingual Reasoning through Test-Time Scaling |
| topic | Computation and Language Artificial Intelligence Machine Learning |
| url | https://arxiv.org/abs/2505.05408 |