aiqbench analysis
This study demonstrates that reinforcement learning can effectively optimize source text rewriting for machine translation, overcoming the limitations of prompt-based methods that often degrade quality with smaller models. By directly optimizing for downstream translation gains, the proposed RLSR framework enables 4B-parameter models to outperform larger baselines, offering a more efficient path for resource-constrained deployment.
Rewrite to Translate, Translate to Reward: Reinforcement Learning for Source Rewriting in Machine Translation
arXiv:2606.08011v2 Announce Type: replace Abstract: Rewriting source text with large language models (LLMs) before translation has been shown to improve machine translation (MT) quality. However, we find that prompt-based rewriting can degrade translation quality rather than improve it, particularly when smaller LLMs, such as 4B-parameter models, are used. We argue that this limitation stems from the difficulty of controlling rewriting behavior through natural-language prompts alone: a rewrite is useful only if it improves downstream translation, yet existing prompt-based methods do not explicitly optimize for this signal. To address this issue, we propose RLSR (Reinforcement Learning for Source Rewriting), a reinforcement learning framework that trains the rewriting model with a reward based on the downstream translation-quality improvement produced by each rewrite. Experiments across six MT systems and 16 language pairs show that our 4B RLSR-trained rewriting models significantly outperform both the no-rewriting baseline and prompt-based rewriting baselines at the same model scale, while remaining competitive with baselines that use a 235B LLM.
Abstract:Rewriting source text with large language models (LLMs) before translation has been shown to improve machine translation (MT) quality. However, we find that prompt-based rewriting can degrade translation quality rather than improve it, particularly when smaller LLMs, such as 4B-parameter models, are used. We argue that this limitation stems from the difficulty of controlling rewriting behavior through natural-language prompts alone: a rewrite is useful only if it improves downstream translation, yet existing prompt-based methods do not explicitly optimize for this signal. To address this issue, we propose RLSR (Reinforcement Learning for Source Rewriting), a reinforcement learning framework that trains the rewriting model with a reward based on the downstream translation-quality improvement produced by each rewrite. Experiments across six MT systems and 16 language pairs show that our 4B RLSR-trained rewriting models significantly outperform both the no-rewriting baseline and prompt-based rewriting baselines at the same model scale, while remaining competitive with baselines that use a 235B LLM.
Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
arXiv:2606.08011 [cs.CL]
(or arXiv:2606.08011v2 [cs.CL] for this version)
https://doi.org/10.48550/arXiv.2606.08011 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 252 words
Related coverage
Rewrite to Translate, Translate to Reward: Reinforcement Learning for Source Rewriting in Machine Translation
arXiv · Jun 9, 2026
Rewrite to Translate, Translate to Reward: Reinforcement Learning for Source Rewriting in Machine Translation
arXiv · Aug 5, 2026
Generating Adversarial Texts for Machine Translation via GRPO
arXiv · Sep 9, 2026
Podcast: The Future of AI: from Enterprise Adoption to Open Source Sovereignty
InfoQ · Sep 25, 2026