Elamur Research⌕
☰
home / papers / 2026 / Can Multi-agent Help Disambiguation in Multi-domain Translation?

Can Multi-agent Help Disambiguation in Multi-domain Translation?

Paper · 2026 · Zhibo Man, Shaoyang Xu, Yujie Zhang et al.

Summary

The paper asks whether LLM-based multi-agent collaboration can resolve cross-domain word ambiguity in multi-domain translation (MDT). Such ambiguity occurs when one source word, like "cell", needs a different rendering depending on the domain, for example law versus science. The authors propose MACD, which splits translation into four prompted roles:

  • a domain allocation agent that labels the input's domain,
  • a general translation agent that produces a meaning-preserving draft,
  • a domain disambiguation agent that produces a domain-conditioned draft,
  • a translation fusion agent that merges both drafts into the final output.

All agents run on Qwen3-30B-A3B-Instruct-2507, with five human-annotated examples per domain. The evaluation covers EN-ZH (UM-Corpus) and a DE-EN multi-domain dataset, with ambiguity test sets taken from DMDTEval. MACD is compared against self-consistency, a three-role multi-agent pipeline, and a single prompt performing all four steps. It achieves the best average BLEU and COMET on both language pairs and higher disambiguation accuracy in every domain. The work shows that splitting domain-aware reasoning across cooperating agents helps more than packing the same steps into one prompt.

Key points

  • Average BLEU/COMET: EN-ZH 41.52/86.22 vs. 40.64/85.63 for the single-prompt SPMS baseline (+0.88 BLEU). DE-EN 33.97/81.46 vs. 33.10/80.48. Metrics are SacreBLEU and COMET (wmt22-comet-da).
  • On EN-ZH, MACD's disambiguation accuracy beats the multi-agent MAT baseline in all five domains (Education, Laws, News, Science, Spoken). The biggest gain is +11.04 points in News.
  • Case studies show that the general and domain agents complement each other. The fusion agent can recover the right term even when the domain label misleads the domain agent.
  • Using only the agent for the assigned domain gives better BLEU and accuracy than adding agents from other domains (tested with 1–5 agents).
  • Grouping 100 same-domain sentences into long inputs raises disambiguation accuracy in every domain, e.g. Laws from 73.14% to 85.66%. The authors see document-level translation as future work.

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

🎓Singapore University of Technology and DesignSingapore · Singapore49 works · 11,074 citations🎓Beijing Jiaotong UniversityBeijing · China22 works · 1,754 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 Zhibo Man, Shaoyang Xu, Yujie Zhang, Yi Feng, Yuanmeng Chen, Yufeng Chen, Xu Jinan and Wenxuan Zhang. 2026. Can Multi-agent Help Disambiguation in Multi-domain Translation?. 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.907

@inproceedings{man2026can,
  title = {{Can Multi-agent Help Disambiguation in Multi-domain Translation?}},
  author = {Zhibo Man, Shaoyang Xu, Yujie Zhang, Yi Feng, Yuanmeng Chen, Yufeng Chen, Xu Jinan and Wenxuan Zhang},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.907},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.907}
}

References

23 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Roee Aharoni and Yoav Goldberg. 2020. Unsupervised domain clusters in pretrained language models. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 7747– 7763, Online. Association for Computational Linguistics.
  2. Zhaopeng Feng, Jiayuan Su, Jiamei Zheng, Jiahan Ren, Yan Zhang, Jian Wu, Hongwei Wang, and Zuozhu Liu. 2025. M-MAD: Multidimensional multi-agent debate for advanced machine translation evaluation. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: L…
  3. Zhiwei He, Tian Liang, Wenxiang Jiao, Zhuosheng Zhang, Yujiu Yang, Rui Wang, Zhaopeng Tu, Shuming Shi, and Xing Wang. 2024. Exploring human-like translation strategy with large language models. Transactions of the Association for Computational Linguistics, 12:229–246.
  4. Tianxiang Hu, Pei Zhang, Baosong Yang, Jun Xie, Derek F. Wong, and Rui Wang. 2024. Large language model for multi-domain translation: Benchmarking and domain CoT fine-tuning. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 5726–5746, Miami, Florida, USA. Association f… find
  5. Haoming Jiang, Chen Liang, Chong Wang, and Tuo Zhao. 2020. Multi-domain neural machine translation with word-level adaptive layer-wise domain mixing. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 1823–1834, Online. Association for Computational Lin… find
  6. Ahrii Kim. 2025. A preliminary study of AI agent model in machine translation. In Proceedings of the Tenth Conference on Machine Translation, pages 583– 586, Suzhou, China. Association for Computational Linguistics. find
  7. Jiyoung Lee, Hantae Kim, Hyunchang Cho, Edward Choi, and Cheonbok Park. 2022. Specializing multi-domain NMT via penalizing low mutual information. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 10015–10026, Abu Dhabi, United Arab Emirates. Associatio… find
  8. Yupu Liang, Yaping Zhang, Zhiyang Zhang, Yang Zhao, Lu Xiang, Chengqing Zong, and Yu Zhou. 2025. Single-to-mix modality alignment with multimodal large language model for document image machine translation. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (V… find
  9. Hongyuan Lu, Zixuan Li, Zefan Zhang, and Wai Lam. 2025. SLoW: Select low-frequency words! automatic dictionary selection for translation on large language models. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 898–913, Suzhou, China. Association for… find
  10. Zhibo Man, Yuanmeng Chen, Yujie Zhang, and Jinan Xu. 2025. DMDTEval: An evaluation and analysis of LLMs on disambiguation in multi-domain translation. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 6065–6082, Suzhou, China. Association for Computatio… find
  11. Zhibo Man, Kaiyu Huang, Yujie Zhang, Yuanmeng Chen, Yufeng Chen, and Jinan Xu. 2024. ICL: Iterative continual learning for multi-domain neural machine translation. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 7732–7743, Miami, Florida, USA. Association for Computat… find
  12. Zhibo Man, Yujie Zhang, Yuanmeng Chen, Yufeng Chen, and Jinan Xu. 2023. Exploring domain-shared and domain-specific knowledge in multi-domain neural machine translation. In Proceedings of Machine Translation Summit XIX, Vol. 1: Research Track, pages 99–110, Macau SAR, China. Asia-Pacific Association… find
  13. Matt Post. 2018. A call for clarity in reporting BLEU scores. In Proceedings of the Third Conference on Machine Translation: Research Papers, pages 186– 191, Brussels, Belgium. Association for Computational Linguistics.
  14. Rui Qi, Zhibo Man, Yufeng Chen, Fengran Mo, Jinan Xu, and Kaiyu Huang. 2025. SoT: Structured-of-thought prompting guides multilingual reasoning in large language models. In Findings of the Association for Computational Linguistics: EMNLP 2025, pages 11024–11039, Suzhou, China. Association for Comput… find
  15. Test set Test set Ricardo Rei, Craig Stewart, Ana C Farinha, and Alon Lavie. 2020. COMET: A neural framework for MT evaluation. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 2685–2702, Online. Association for Computational Linguistics.
  16. Liang Tian, Derek F. Wong, Lidia S. Chao, Paulo Quaresma, Francisco Oliveira, Yi Lu, Shuo Li, Yiming Wang, and Longyue Wang. 2014. UM-corpus: A large English-Chinese parallel corpus for statistical machine translation. In Proceedings of the Ninth International Conference on Language Resources and Ev… find
  17. Minghao Wu, Jiahao Xu, and Longyue Wang. 2024. TransAgents: Build your translation company with language agents. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, pages 131–141, Miami, Florida, USA. Association for Computational Linguis… find
  18. Minghao Wu, Jiahao Xu, Yulin Yuan, Gholamreza Haffari, Longyue Wan, Weihua Luo, and Kaifu Zhang. 2025. (perhaps) beyond human translation: Harnessing multi-agent collaboration for translating ultra-long literary texts. Transactions of the Association for Computational Linguistics, 13:901–922.
  19. Yongshi Ye, Biao Fu, Chongxuan Huang, Yidong Chen, and Xiaodong Shi. 2025. How well do large reasoning models translate? a comprehensive evaluation for multi-domain machine translation. arXiv preprint arXiv:2505.19987. find
  20. Sub 2000 *Table 4: The statistics of MDT test sets. Edu represents for the Education domain, Sci represents for the Science domain, Spo represents for the Spoken domain, and Sub represents for the subtitles domain.* Rui Qi, Fengran Mo, Yufeng Chen, Xue Zhang, Shuo Wang, Hongliang Li, Jinan Xu, Meng… — [arXiv:2601.14896](https://arxiv.org/abs/2601.14896) find
  21. Shenbin Qian, Archchana Sindhujan, Minnie Kabra, Diptesh Kanojia, Constantin Orasan, Tharindu Ranasinghe, and Fred Blain. 2024. What do large language models need for machine translation evaluation? In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 3660… find
  22. English-to-Chinese Edu Laws News Sci 790 456 1500 503 German-to-English IT Kor Laws Med 2000 2000 2000 2000 find
  23. Test set Test set English-to-Chinese Edu Laws News Sci 492 696 720 471 German-to-English IT Kor Laws Med 214 167 412 345 Spo 432 Sub 209 *Table 5: Statistics of the number of cross-domain ambiguous words on MDT* find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/man-2026-can-multi-agent-help-disambiguation/