[2508.14828] Long Chain-of-Thought Reasoning Across Languages

Computer Science > Computation and Language arXiv:2508.14828 (cs) [Submitted on 20 Aug 2025 (v1), last revised 20 Mar 2026 (this version, v3)] Title:Long Chain-of-Thought Reasoning Across Languages Authors:Josh Barua, Seun Eisape, Kayo Yin, Alane Suhr View a PDF of the paper titled Long Chain-of-Thought Reasoning Across Languages, by Josh Barua and 3 other authors View PDF HTML (experimental) Abstract:While large reasoning models have shown remarkable ability to generate long chains-of-thought (CoTs) in English, we still lack understanding of how these long-form reasoning abilities transfer to the vast majority of the world's languages. In this work, we systematically investigate four key stages of model development--scaling, pretraining, post-training, and inference--to understand how long CoT capabilities extend beyond English. We compare two reasoning settings across nine non-English target languages: En-CoT, where models process target-language inputs, but reason in English; and Target-CoT, where models both process inputs and generate long CoTs in the target language. We find that scaling reasoning model size improves multilingual task performance in En-CoT, but Target-CoT performance lags behind. This gap widens for tasks requiring long, multi-step CoTs such as mathematical reasoning. Shifting to pretraining, we find that adding a specialized reasoning stage enhances En-CoT performance but degrades Target-CoT, whereas broad multilingual pretraining improves both mode...