Elamur Research⌕
☰
home / papers / 2026 / Bootstrapping Code Translation with Weighted Multilanguage Exploration

Bootstrapping Code Translation with Weighted Multilanguage Exploration

Paper · 2026 · Yuhan Wu, Huan Zhang, Wei Cheng et al.

Summary

Translating code between several programming languages with LLMs runs into two problems. Parallel code that comes with executable tests is scarce, and training on many language pairs at once tends to favor the easier directions. BootTrans tackles both with reinforcement learning from verifiable rewards. It starts from Python code that already has unit tests and converts those tests by rule into other languages, so they act as shared correctness checks. Training uses two data pools. A seed pool holds the Python code-test pairs. An exploration pool collects the model's own translations that pass the tests, and these become new source inputs, which opens up translation directions that do not start from Python. A language-aware weighting scheme then increases the learning signal for a target language when the model does poorly there but succeeds on sibling languages for the same input. Across C++, Java and Python, BootTrans beats its base models on HumanEval-X and TransCoder-Test. This matters because it shows multilingual translation can be trained without parallel corpora.

Key points

  • Unit tests are moved from the pivot language (Python) to other languages with rule-based conversion, using the MultiPL-E translator. The reward is binary: it is 1 only if the output compiles and passes every test, while errors and timeouts score 0.
  • The exploration pool is a FIFO queue whose capacity equals the number of non-pivot languages times the rollout batch size. Only verified rollouts that originate from seed examples are added, which limits semantic drift. Training batches are drawn from this pool first.
  • Weighted GRPO adjusts the loss for each translation direction according to how well the model does on that target language compared with sibling languages.
  • The largest gains over the base model are up to 26.82% on HumanEval-X and 7.46% on TransCoder-Test. Results on ClassEval-T are also reported for Qwen3-1.7B and Qwen2.5-7B-Instruct, and some directions there stay close to zero.
  • The authors list these limitations:
    • Only imperative languages were tested.
    • Binary rewards do not capture readability or style.
    • Performance depends on the size of the test suite.
  • Code: github.com/nju-websoft/BootTrans

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

🎓Nanjing UniversityNanjing · China99 works · 8,510 citations

 Fields, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksInformation Systems · 1,730 worksSoftware · 119 works

Cite

Cite as Yuhan Wu, Huan Zhang, Wei Cheng, Chen Shen, Jingyue Yang and Wei Hu. 2026. Bootstrapping Code Translation with Weighted Multilanguage Exploration. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-long.1678

@inproceedings{wu2026bootstrapping,
  title = {{Bootstrapping Code Translation with Weighted Multilanguage Exploration}},
  author = {Yuhan Wu, Huan Zhang, Wei Cheng, Chen Shen, Jingyue Yang and Wei Hu},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-long.1678},
  url = {https://doi.org/10.18653/v1/2026.acl-long.1678}
}

References

37 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Federico Cassano, John Gouwar, Francesca Lucchetti, Claire Schlesinger, Anders Freeman, Carolyn Jane Anderson, Molly Q Feldman, Michael Greenberg, Abhinav Jangda, and Arjun Guha. 2024. Knowledge transfer from high-resource to low-resource programming languages for code LLMs. In OOPSLA2, pages 677–70… find
  2. Federico Cassano, John Gouwar, Daniel Nguyen, Sydney Nguyen, Luna Phipps-Costin, Donald Pinckney, Ming-Ho Yee, Yangtian Zi, Carolyn Jane Anderson, Molly Q Feldman, Arjun Guha, Michael Greenberg, and Abhinav Jangda. 2023. MultiPL-E: A scalable and extensible approach to benchmarking neural code gener…
  3. Yali Du, Hui Sun, and Ming Li. 2024. A joint learning model with variational interaction for multilingual program translation. In ASE, pages 1907–1918, Sacramento, CA, USA. ACM. find
  4. Hasan Ferit Enişer, Valentin Wüstholz, and Maria Christakis. 2024. Automatically testing functional properties of code translation models. In AAAI, volume 38, pages 21055–21062. find
  5. Zhangyin Feng, Daya Guo, Duyu Tang, Nan Duan, Xiaocheng Feng, Ming Gong, Linjun Shou, Bing Qin, Ting Liu, Daxin Jiang, and Ming Zhou. 2020. Code-BERT: A pre-trained model for programming and natural languages. In EMNLP, pages 1536–1547, Virtual. ACL. find
  6. Daya Guo, Shuo Ren, Shuai Lu, Zhangyin Feng, Duyu Tang, Shujie Liu, Long Zhou, Nan Duan, Alexey Svyatkovskiy, Shengyu Fu, Michele Tufano, Shao Kun Deng, Colin B. Clement, Dawn Drain, Neel Sundaresan, Jian Yin, Daxin Jiang, and Ming Zhou. 2021. GraphCodeBERT: Pre-training code representations with da…
  7. Yufan Huang, Mengnan Qi, Yongqiang Yao, Maoquan Wang, Bin Gu, Colin Clement, and Neel Sundaresan. 2023. Program translation via code distillation. In EMNLP, pages 10903–10914, Singapore. ACL. find
  8. Prithwish Jana, Piyush Jha, Haoyang Ju, Gautham Kishore, Aryan Mahajan, and Vijay Ganesh. 2024. CoTran: An LLM-based code translator using reinforcement learning with feedback from compiler and symbolic execution. In ECAI, pages 1–15, Santiago de Compostela, Spain. IOS Press. find
  9. Mingsheng Jiao, Tingrui Yu, Xuan Li, Guanjie Qiu, Xiaodong Gu, and Beijun Shen. 2023. On the evaluation of neural code translation: Taxonomy and benchmark. In ASE, pages 1529–1541, Kirchberg, Luxembourg. IEEE. find
  10. Mohammad Abdullah Matin Khan, M. Saiful Bari, Xuan Do Long, Weishi Wang, Md. Rizwan Parvez, and Shafiq Joty. 2024. XCodeEval: An execution-based large scale multilingual multitask benchmark for code understanding, generation, translation and retrieval. In ACL, pages 6766–6805, Bangkok, Thailand. ACL…
  11. Hung Le, Yue Wang, Akhilesh Deepak Gotmare, Silvio Savarese, and Steven C.H. Hoi. 2022. CodeRL: Mastering code generation through pretrained models and deep reinforcement learning. In NeurIPS, pages 1–15, New Orleans, LA, USA. Curran Associates, Inc.
  12. Fang Liu, Jia Li, and Li Zhang. 2023. Syntax and domain aware model for unsupervised program translation. In ICSE, pages 755–767, Melbourne, VIC, Australia. IEEE. find
  13. Marcos Macedo, Yuan Tian, Pengyu Nie, Filipe R. Cogo, and Bram Adams. 2025. InterTrans: Leveraging transitive intermediate translations to enhance LLM-based code translation. In ICSE, pages 1153– 1164, Ottawa, ON, Canada. IEEE. find
  14. Anh Tuan Nguyen, Tung Thanh Nguyen, and Tien N. Nguyen. 2014. Migrating code with statistical machine translation. In ICSE Companion, pages 544– 547, Hyderabad, India. ACM. find
  15. Rosel Oida-Onesa and Melvin A. Ballera. 2024. Fine tuning language models: A tale of two low-resource languages. Data Intelligence, 6(4):946–967. find
  16. Rangeet Pan, Ali Reza Ibrahimzada, Rahul Krishna, Divya Sankar, Lambert Pouguem Wassi, Michele Merler, Boris Sobolev, Raju Pavuluri, Saurabh Sinha, and Reyhaneh Jabbarvand. 2024. Lost in translation: A study of bugs introduced by large language models while translating code. In ICSE, pages 1–13. find
  17. Baptiste Roziere, Marie-Anne Lachaux, Lowik Chanussot, and Guillaume Lample. 2020. Unsupervised translation of programming languages. In NeurIPS, pages 20601–20611, Vancouver, BC, Canada. Curran Associates Inc.
  18. Baptiste Roziere, Jie Zhang, Francois Charton, Mark Harman, Gabriel Synnaeve, and Guillaume Lample. 2022. Leveraging automated unit tests for unsupervised code translation. In ICLR, pages 1–20, Virtual. OpenReview.net. find
  19. John Schulman, Filip Wolski, Prafulla Dhariwal, Alec Radford, and Oleg Klimov. 2017. Proximal policy optimization algorithms. CoRR, 1707.06347:1–12. find
  20. Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, Y.K. Li, Y. Wu, and Daya Guo. 2024. DeepSeekMath: Pushing the limits of mathematical reasoning in open language models. CoRR, 2402.03300:1–30.
  21. Parshin Shojaee, Aneesh Jain, Sindhu Tipirneni, and Chandan K. Reddy. 2023. Execution-based code generation using deep reinforcement learning. Trans. Mach. Learn. Res., 2023:1–26. find
  22. Marc Szafraniec, Baptiste Roziere, Hugh Leather Francois Charton, Patrick Labatut, and Gabriel Synnaeve. 2023. Code translation with compiler representations. In ICLR, pages 1–20, Kigali, Rwanda. Open-Review.net. find
  23. Zilu Tang, Mayank Agarwal, Alexander Shypula, Bailin Wang, Derry Wijaya, Jie Chen, and Yoon Kim. 2023. Explain-then-translate: an analysis on improving program translation with self-generated explanations. In EMNLP Findings, pages 1741–1788. find
  24. Yanli Wang, Yanlin Wang, Suiquan Wang, Daya Guo, Jiachi Chen, John Grundy, Xilin Liu, Yuchi Ma, Mingzhi Mao, Hongyu Zhang, and Zibin Zheng. 2024. RepoTransBench: A real-world benchmark for repository-level code translation. CoRR, 2412.17744:1–23. find
  25. Yanlin Wang, Rongyi Ou, Yanli Wang, Mingwei Liu, Jiachi Chen, Ensheng Shi, Xilin Liu, Yuchi Ma, and Zibin Zheng. 2025. EffiReasonTrans: RL-optimized reasoning for code translation. arXiv, 2510.18863:1– 30. find
  26. Yue Wang, Weishi Wang, Shafiq Joty, and Steven C.H. Hoi. 2021. CodeT5: Identifier-aware unified pre-trained encoder-decoder models for code understanding and generation. In EMNLP, pages 8696–8708, Punta Cana, Dominican Republic. ACL. find
  27. Haoyuan Wu, Rui Ming, Jilong Gao, Hangyu Zhao, Xueyi Chen, Yikai Yang, Haisheng Zheng, Zhuolun He, and Bei Yu. 2025. On-policy optimization with group equivalent preference for multi-programming language understanding. In NeurIPS, pages 1–38, Sydney, Australia. OpenReview.net. find
  28. Zhangchen Xu, Yang Liu, Yueqin Yin, Mingyuan Zhou, and Radha Poovendran. 2025. KodCode: A diverse, challenging, and verifiable synthetic dataset for coding. In ACL Findings, pages 6980–7008, Vienna, Austria. ACL. find
  29. Pengyu Xue, Linhao Wu, Zhen Yang, Chengyi Wang, Xiang Li, Yuxiang Zhang, Jia Li, Ruikai Jin, Yifei Pei, Zhaoyan Shen, and 1 others. 2025. ClassEval-T: Evaluating large language models in class-level code translation. ISTTA, 2:1421–1444. find
  30. Weixiang Yan, Haitian Liu, Yunkun Wang, Yunzhe Li, Qian Chen, Wen Wang, Tingyu Lin, Weishan Zhao, Li Zhu, Hari Sundaram, and Shuiguang Deng. 2024. CodeScope: An execution-based multilingual multitask multidimensional benchmark for evaluating LLMs on code understanding and generation. In ACL, pages 5… find
  31. Weixiang Yan, Yuchen Tian, Yunzhe Li, Qian Chen, and Wen Wang. 2023. CodeTransOcean: A comprehensive multilingual benchmark for code translation. In EMNLP, pages 5067–5089, Singapore. ACL. find
  32. Zhen Yang, Fang Liu, Zhongxing Yu, Jacky Wai Keung, Jia Li, Shuo Liu, Yifan Hong, Xiaoxue Ma, Zhi Jin, and Ge Li. 2024. Exploring and unleashing the power of large language models in automated code translation. Proc. ACM Softw. Eng., 1:1585–1608. find
  33. Jiyang Zhang, Pengyu Nie, Junyi Jessy Li, and Milos Gligoric. 2023. Multilingual code co-evolution using large language models. In FSE, pages 695–707. find
  34. Qinkai Zheng, Xiao Xia, Xu Zou, Yuxiao Dong, Shan Wang, Yufei Xue, Lei Shen, Zihan Wang, Andi Wang, Yang Li, Teng Su, Zhilin Yang, and Jie Tang. 2023. CodeGeeX: A pre-trained model for code generation with multilingual benchmarking on HumanEval-X. In KDD, pages 5673–5684, Long Beach, CA, USA. ACM.
  35. Ming Zhu, Aneesh Jain, Karthik Suresh, Roshan Ravindran, Sindhu Tipirneni, and Chandan K. Reddy. 2022a. XLCoST: A benchmark dataset for cross-lingual code intelligence. CoRR, 2206.08474:1–20. find
  36. Ming Zhu, Mohimenul Karim, Ismini Lourentzou, and Daphne Yao. 2024. Semi-supervised code translation overcoming the scarcity of parallel code data. In ASE, pages 1545–1556, Sacramento, CA, USA. IEEE/ACM. find
  37. Ming Zhu, Karthik Suresh, and Chandan K. Reddy. 2022b. Multilingual code snippets training for program translation. In AAAI, volume 36, pages 11783– 11790, Virtual. AAAI. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wu-2026-bootstrapping-code-translation-weighted-multilanguage/