CoTranslator: Reverse to Forward Chain-of-Thought Translator for Reasoning Distillation
Quick Answer
This paper shows that CoTranslator efficiently translates reverse chain-of-thought responses to verified forward responses, improving average accuracy by up to 5.82 percentage points across six math benchmarks.
Quick Take
It retains 95% of rejection sampling's accuracy while using only 35.9% of its compute, demonstrating transferability to new datasets and problems.
Key Points
- CoTranslator improves accuracy over semantic suppression by up to 5.82 percentage points.
- Retains 95% accuracy of rejection sampling while using only 35.9% of compute.
- Demonstrates transferability to new math datasets and different teacher models.
- Improves average STEM accuracy by 4.88 and 4.31 percentage points for 4B and 8B students.
- Trained on a small set of paired reverse and rejection-sampled forward responses.
Paper Resources
📖 Reader Mode
~2 min readAbstract:Reasoning distillation requires chain-of-thought (CoT) data with correct answers, but obtaining such data efficiently remains challenging. Forward CoT generation (FCG) followed by rejection sampling (RS) produces verified, question-only responses, yet may require many teacher calls for difficult problems. Reverse CoT generation (RCG) is cheaper because it conditions on the known answer, but access to this answer can lead to post-hoc rationalization and a mismatch with the student's answer-free inference setting. We propose CoTranslator, which learns to translate reverse CoT responses toward the distribution of verified forward responses. Trained on a small set of paired reverse and rejection-sampled forward responses, it supports large-scale generation and verification of student training data. Across six student models evaluated on six mathematics benchmarks, CoTranslator improves average accuracy over semantic suppression by up to $5.82$ percentage points. For two 4B students, it retains $95.0\%$ of rejection sampling's average accuracy while using $35.9\%$ of its estimated compute. Without retraining, the same translator transfers to a new mathematics dataset, a different teacher, and STEM problems, improving average STEM accuracy by $4.88$ and $4.31$ percentage points for 4B and 8B students, respectively. These results show that CoTranslator offers an efficient and transferable approach to reasoning distillation.
| Comments: | 18 pages, 12 figures, under review |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2610.20828 [cs.CL] |
| (or arXiv:2610.20828v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.20828 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Zinan Tang [view email]
[v1]
Wed, 14 Oct 2026 19:15:36 UTC (419 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CL
See more →The "10th Juror": Open-Set Standpoint Screening for Bureaucratic Bias Detection
MARS-Gov introduces a framework for detecting bureaucratic bias in Dutch government documents, achieving a new state-of-the-art F1 score of 0.880. This model outperforms existing zero-shot detectors by 20.2 points and reduces unnecessary interventions to just 2.5%. The framework's dynamic '10th juror' adapts to emerging biases, enhancing legal language processing.