Elamur Research⌕
☰
home / papers / 2026 / Learning to Translate by Translating: Stabilizing the Dual Loop via Semantic-Aware Self-Evolution

Learning to Translate by Translating: Stabilizing the Dual Loop via Semantic-Aware Self-Evolution

Paper · 2026 · Kui Liu, Mingming Yin, Zuoli Tang et al.

Summary

Supervised LLM translation systems such as Tower and X-ALMA rely on large, expensive human-made parallel corpora. This work asks whether an LLM can improve its translation ability using only monolingual text. The authors present LTT (Learning to Translate by Translating), which revives classic dual learning for generative models. A single policy model translates a sentence into the target language and then back into the source language. The round trip supplies a training signal, and the model is optimized with GRPO.

Earlier unsupervised MT often suffered from reward hacking, for example copying the source or producing overly literal output. To prevent this, the reward combines several parts:

  • a reference-free adequacy score (COMETkiwi) for the forward translation;
  • a reconstruction score (BLEU) for the back-translation;
  • penalties for copying the source and for mixing languages;
  • a formatting check.

The resulting 8B model performs close to supervised systems on WMT EN-ZH and beats much larger models in multilingual and low-resource settings. This makes it a cheap, data-efficient path for translation in under-resourced language pairs.

Key points

  • Reward design: the final reward is 1 + R_Q + R_C for well-formed outputs and −3 for outputs that break the <translate> tag format.
    • R_Q = COMETkiwi(x, ŷ) + BLEU(x̂, x).
    • R_C = −BLEU(x, ŷ), plus a −0.5 penalty when language mixing is detected.
  • EN-ZH results: the 8B model scores 77.58 in semantic evaluation. This is slightly above Gemini-2.5-Pro (77.55) and above Qwen2.5-72B-Instruct (76.52). It trails the best supervised baseline by only 0.07%.
  • Multilingual results: the score rises from 32.22 (base model) to 58.51, ahead of Gemma2-9B-it and LLaMA-3.1-70B. The paper reports an overall 80.42% gain over the base model.
  • Ablations:
    • BLEU-only rewards lead to stiff word-for-word translations that fall below the base model.
    • COMET-only rewards produce fluent but loosely faithful output.
    • Removing the copy penalty causes source words to leak into translations.
    • Removing the mixing penalty sometimes yields output in the wrong target language.
  • Limitations: training is costlier than supervised fine-tuning because multiple candidates must be sampled per input. Quality is also capped by what the base LLM learned in pre-training, so languages absent from that data cannot be bootstrapped.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓Wuhan UniversityWuhan · China35 works · 1,127 citations🏢Ant GroupHangzhou · China42 works · 1,131 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

Cite

Cite as Kui Liu, Mingming Yin, Zuoli Tang, Zihao Li, Chilin Fu, Xiaolu Zhang, Jun Zhou, Lixin Zou and Chenliang Li. 2026. Learning to Translate by Translating: Stabilizing the Dual Loop via Semantic-Aware Self-Evolution. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.723

@inproceedings{liu2026learning,
  title = {{Learning to Translate by Translating: Stabilizing the Dual Loop via Semantic-Aware Self-Evolution}},
  author = {Kui Liu, Mingming Yin, Zuoli Tang, Zihao Li, Chilin Fu, Xiaolu Zhang, Jun Zhou, Lixin Zou and Chenliang Li},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.723},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.723}
}

References

35 references, 14 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Sweta Agrawal, Chunting Zhou, Mike Lewis, Luke Zettlemoyer, and Marjan Ghazvininejad. 2023. In-context examples selection for machine translation. In Findings of the Association for Computational Linguistics: ACL 2023, pages 8857–8873.
  2. Duarte M Alves, José Pombal, Nuno M Guerreiro, Pedro H Martins, João Alves, Amin Farajian, Ben Peters, Ricardo Rei, Patrick Fernandes, Sweta Agrawal, Duarte Miguel Alves, José Pombal, Nuno M Guerreiro, Pedro Henrique Martins, João Alves, Amin Farajian, Ben Peters, Ricardo Rei, Patrick Fernandes, Swe… — PaLM 2 Technical Report
  3. Mauro Cettolo, Marcello Federico, Luisa Bentivogli, Jan Niehues, Sebastian Stüker, Katsuhito Sudoh, Koichiro Yoshino, and Christian Federmann. 2017. Overview of the iwslt 2017 evaluation campaign. In Proceedings of the 14th International Conference on Spoken Language Translation, pages 2–14.
  4. Gheorghe Comanici, Eric Bieber, Mike Schaekermann, Ice Pasupat, Noveen Sachdeva, Inderjit Dhillon, Marcel Blistein, Ori Ram, Dan Zhang, Evan Rosen, and 1 others. 2025. Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities. arX…
  5. Marta R Costa-Jussà, James Cross, Onur Çelebi, Maha Elbayad, Kenneth Heafield, Kevin Heffernan, Elahe Kalbassi, Janice Lam, Daniel Licht, Jean Maillard, and 1 others. 2022. No language left behind: Scaling human-centered machine translation. arXiv preprint arXiv:2207.04672. find
  6. Menglong Cui, Pengzhi Gao, Wei Liu, Jian Luan, and Bin Wang. 2025. Multilingual machine translation with open large language models at practical scale: An empirical study. In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: H… find
  7. Zhaopeng Feng, Shaosheng Cao, Jiahan Ren, Jiayuan Su, Ruizhe Chen, Yan Zhang, Zhe Xu, Yao Hu, Jian Wu, and Zuozhu Liu. 2025. Mt-r1-zero: Advancing llm-based machine translation via r1-zero-like reinforcement learning. arXiv preprint arXiv:2504.10160. find
  8. Zhaopeng Feng, Yan Zhang, Hao Li, Wenqiang Liu, Jun Lang, Yang Feng, Jian Wu, and Zuozhu Liu. 2024. Improving llm-based machine translation with systematic self-correction. CoRR. find
  9. Pengzhi Gao, Zhongjun He, Hua Wu, and Haifeng Wang. 2024. Towards boosting many-to-many multilingual machine translation with large language models. arXiv preprint arXiv:2401.05861. — [arXiv:2401.05861](https://arxiv.org/abs/2401.05861) find
  10. Nuno M Guerreiro, Ricardo Rei, Daan van Stigt, Luisa Coheur, Pierre Colombo, and André FT Martins. 2024. xcomet: Transparent machine translation evaluation through fine-grained error detection. Transactions of the Association for Computational Linguistics, 12:979–995.
  11. Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, and 1 others. 2025. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning. arXiv preprint arXiv:2501.12948.
  12. Jiaxin Guo, Hao Yang, Zongyao Li, Daimeng Wei, Hengchao Shang, and Xiaoyu Chen. 2024. A novel paradigm boosting translation capabilities of large language models. In Findings of the Association for Computational Linguistics: NAACL 2024, pages 639–649. find
  13. Di He, Yingce Xia, Tao Qin, Liwei Wang, Nenghai Yu, Tie-Yan Liu, and Wei-Ying Ma. 2016. Dual learning for machine translation. Advances in neural information processing systems, 29. find
  14. Minggui He, Yilun Liu, Shimin Tao, Yuanchang Luo, Hongyong Zeng, Chang Su, Li Zhang, Hongxia Ma, Daimeng Wei, Weibin Meng, and 1 others. 2025. R1-t1: Fully incentivizing translation capability in llms via reasoning learning. arXiv preprint arXiv:2502.19735.
  15. Aaron Hurst, Adam Lerer, Adam P Goucher, Adam Perelman, Aditya Ramesh, Aidan Clark, AJ Ostrow, Akila Welihinda, Alan Hayes, Alec Radford, and 1 others. 2024. Gpt-4o system card. arXiv preprint arXiv:2410.21276.
  16. Aaron Jaech, Adam Kalai, Adam Lerer, Adam Richardson, Ahmed El-Kishky, Aiden Low, Alec Helyar, Aleksander Madry, Alex Beutel, Alex Carney, and 1 others. 2024. Openai o1 system card. arXiv preprint arXiv:2412.16720.
  17. Tom Kocmi and Christian Federmann. 2023. Large language models are state-of-the-art evaluators of translation quality. In 24th Annual Conference of the European Association for Machine Translation, page 193. find
  18. Guillaume Lample, Alexis Conneau, Ludovic Denoyer, and Marc’Aurelio Ranzato. 2018. Unsupervised machine translation using monolingual corpora only. In International Conference on Learning Representations.
  19. Chong Li, Shaonan Wang, Jiajun Zhang, and Chengqing Zong. 2024. Improving in-context learning of multilingual generative language models with cross-lingual alignment. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language… find
  20. Kelly Marchisio, Kevin Duh, and Philipp Koehn. 2020. When does unsupervised machine translation work? In Proceedings of the Fifth Conference on Machine Translation, pages 571–583. find
  21. Mariana Neves, Cristian Grozea, Philippe Thomas, Roland Roller, Rachel Bawden, Aurélie Névéol, Steffen Castle, Vanessa Bonato, Giorgio Maria Di Nunzio, Federica Vezzani, and 1 others. 2024. Findings of the wmt 2024 biomedical translation shared task: Test sets on abstract level. In Proceedings of th… find
  22. Marc’Aurelio Ranzato, Sumit Chopra, Michael Auli, and Wojciech Zaremba. 2016. Sequence level training with recurrent neural networks. In 4th International Conference on Learning Representations, ICLR 2016.
  23. Ricardo Rei, Nuno M Guerreiro, José Pombal, João Alves, Pedro Teixeirinha, Amin Farajian, and André FT Martins. 2025. Tower+: Bridging generality and translation specialization in multilingual llms. arXiv preprint arXiv:2506.17080. find
  24. Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, YK Li, Yang Wu, and 1 others. 2024. Deepseekmath: Pushing the limits of mathematical reasoning in open language models. arXiv preprint arXiv:2402.03300.
  25. Gemma Team, Morgane Riviere, Shreya Pathak, Pier Giuseppe Sessa, Cassidy Hardin, Surya Bhupatiraju, Léonard Hussenot, Thomas Mesnard, Bobak Shahriari, Alexandre Ramé, and 1 others. 2024. Gemma 2: Improving open language models at a practical size. arXiv preprint arXiv:2408.00118. find
  26. Jiaan Wang, Fandong Meng, Yunlong Liang, and Jie Zhou. 2024. Drt-o1: Optimized deep reasoning translation via long chain-of-thought. arXiv e-prints, pages arXiv–2412. find
  27. Jiaan Wang, Fandong Meng, and Jie Zhou. 2025. Deep reasoning translation via reinforcement learning. arXiv preprint arXiv:2504.10187. — [arXiv:2504.10187](https://arxiv.org/abs/2504.10187) find
  28. Lijun Wu, Li Zhao, Tao Qin, Jianhuang Lai, and Tie-Yan Liu. 2017. Sequence prediction with unlabeled data by reward function learning. In IJCAI, pages 3098–3104. find
  29. Tianhao Wu, Weizhe Yuan, Olga Golovneva, Jing Xu, Yuandong Tian, Jiantao Jiao, Jason Weston, and Sainbayar Sukhbaatar. 2024. Meta-rewarding language models: Self-improving alignment with llm-as-ameta-judge. arXiv preprint arXiv:2407.19594. — [arXiv:2407.19594](https://arxiv.org/abs/2407.19594) find
  30. Haoran Xu, Young Jin Kim, Amr Sharaf, and Hany Hassan Awadalla. a. A paradigm shift in machine translation: Boosting translation performance of large language models. In The Twelfth International Conference on Learning Representations. Haoran Xu, Kenton Murray, Philipp Koehn, Hieu Hoang, Akiko Erigu… — Qwen3 Technical Report
  31. An Yang, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chengyuan Li, Dayiheng Liu, Fei Huang, Haoran Wei, and 1 others. 2024. Qwen2.5 technical report. arXiv e-prints, pages arXiv–2412. find
  32. Weizhe Yuan, Richard Yuanzhe Pang, Kyunghyun Cho, Xian Li, Sainbayar Sukhbaatar, Jing Xu, and Jason Weston. 2024. Self-rewarding language models. In Proceedings of the 41st International Conference on Machine Learning, pages 57905–57923. find
  33. Yu Zhao, Huifeng Yin, Bo Zeng, Hao Wang, Tianqi Shi, Chenyang Lyu, Longyue Wang, Weihua Luo, and Kaifu Zhang. 2024. Marco-o1: Towards open reasoning models for open-ended solutions. arXiv preprint arXiv:2411.14405. find
  34. Wenhao Zhu, Hongyi Liu, Qingxiu Dong, Jingjing Xu, Shujian Huang, Lingpeng Kong, Jiajun Chen, and Lei Li. 2024. Multilingual machine translation with large language models: Empirical results and analysis. In Findings of the Association for Computational Linguistics: NAACL 2024, pages 2765–2781.
  35. Wei Zou, Sen Yang, Yu Bao, Shujian Huang, Jiajun Chen, and Shanbo Cheng. 2025. Trans-zero: Self-play incentivizes large language models for multilingual translation without parallel data. arXiv preprint arXiv:2504.14669. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/liu-2026-learning-translate-translating-stabilizing-dual/