Elamur Research⌕
☰
home / papers / 2026 / Multi-Agent Orchestration for Terminology-Constrained Machine Translation in Industrial Localization

Multi-Agent Orchestration for Terminology-Constrained Machine Translation in Industrial Localization

Paper · 2026 · Emanuele Di Rosa

Summary

In industrial localization, enterprise clients require approved glossary terms to appear exactly in translations, and one wrong term can breach a contract. This paper describes AIDA term, a production system at DATAmundi.ai that chains four LLM agents with distinct roles: Analysis, Translation, Post-editing and Review. The Analysis agent profiles domain, audience and register to help resolve ambiguous terms. Glossary entries are supplied to every agent, the Translation agent uses a different model family (Claude 4.5 Sonnet) from the GPT-5/5.2 agents, and the Review agent can use high reasoning effort. The author compares six configurations on WMT25 Terminology Track 1 (IT domain, en→de/es/ru), using the official scripts after the shared task had closed. The strongest variant, V5, reaches 99.4% average terminology accuracy and the top ChrF2++ for all three pairs, ranking above all 20 submitted systems. The ablation suggests that the pipeline design matters more than the choice of model. Production human evaluations indicate that much of the output needs little or no editing.

Key points

  • Configurations: V1, V1.1, V2, V3, V4 and V5 vary the Analysis agent, strict term directives, batch versus per-segment terms, model generation and review reasoning. Five of the six exceed every WMT25 submission on the official ChrF2++×Acc metric.
  • Older models: V4 uses GPT-4o and Claude 3.7 Sonnet, both released before the test set, and no reasoning. It still beats the top single-agent system, o3-term-guide, with a fixed number of API calls.
  • Human evaluation against Microsoft Translator: segments rated Good or Excellent rose from 82.7% to 98.0% in German and from 12.2% to 69.4% in Japanese.
  • Production evaluation: 4,418 English→Polish engineering and science segments went through post-editing and then independent review. The first translator left 42.4% unchanged, and the reviewer left 34.9% unchanged.
  • Deployment and resources: the system works natively with XLIFF for CAT tool integration. Prompts and output translations are released on GitHub (emanueledirosa/aida_t-acl2026-industrytrack).

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

 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

🧭 Topics 1Research subjects and languages the work is about.Machine translation · 1,087 works

Cite

Cite as Emanuele Di Rosa. 2026. Multi-Agent Orchestration for Terminology-Constrained Machine Translation in Industrial Localization. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.acl-industry.63

@inproceedings{rosa2026multi,
  title = {{Multi-Agent Orchestration for Terminology-Constrained Machine Translation in Industrial Localization}},
  author = {Emanuele {Di Rosa}},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 6: Industry Track)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-industry.63},
  url = {https://doi.org/10.18653/v1/2026.acl-industry.63}
}

References

21 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Md Mahfuz Ibn Alam, Antonios Anastasopoulos, Laurent Besacier, and 1 others. 2021. Findings of the WMT 2021 shared task on machine translation using terminologies. In Proceedings of the Sixth Conference on Machine Translation (WMT). Association for Computational Linguistics. find
  2. V. Briva-Iglesias, C. Campolargo, S. Coyne, and A. Way. 2024. Legal translation: The potential of multi-agent LLMs for improved efficiency and quality. In Proceedings of the 16th Conference of the European Association for Machine Translation, pages 421–430. European Association for Machine Translati… find
  3. Georgiana Dinu, Prashant Mathur, Marcello Federico, and Yaser Al-Onaizan. 2019. Training neural machine translation to apply terminology constraints. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 3063– 3068. Association for Computational Linguistic…
  4. Miriam Exel, Bianka Buschbeck, Laurie Brandt, and Simona Doneva. 2020. Terminology-constrained neural machine translation at SAP. In Proceedings of the 22nd Annual Conference of the European Association for Machine Translation, pages 271–280. find
  5. Javier Garcia Gilabert, Carlos Escolano, Xixian Liao, and Maite Melero. 2025. Terminology-constrained translation from monolingual data using GRPO. In Proceedings of the Tenth Conference on Machine Translation (WMT), pages 1335–1343, Suzhou, China. Association for Computational Linguistics. find
  6. Ivan Grubišić and Damir Korenčić. 2025. IRB-MT at WMT25 terminology translation task: Metric-guided multi-agent approach. In Proceedings of the Tenth Conference on Machine Translation (WMT), pages 1302–1334, Suzhou, China. Association for Computational Linguistics. find
  7. Amr Hendy, Mohamed Abdelrehim, Amr Sharaf, and 1 others. 2023. How good are GPT models at machine translation? A comprehensive evaluation. arXiv preprint arXiv:2302.09210. find
  8. Akshat Singh Jaswal. 2025. It takes two: A dual stage approach for terminology-aware translation. In Proceedings of the Tenth Conference on Machine Translation (WMT), Suzhou, China. Association for Computational Linguistics. find
  9. Wenxiang Jiao, Wenxuan Wang, Jen-tse Huang, Xing Wang, Shuming Shi, and Zhaopeng Tu. 2023. Is ChatGPT a good translator? Yes with GPT-4 as the engine. arXiv preprint arXiv:2301.08745.
  10. Tom Kocmi, Arkady Arkhangorodsky, Alexandre Berard, and 1 others. 2025a. Command-A-translate: Raising the bar of machine translation with difficulty filtering. In Proceedings of the Tenth Conference on Machine Translation (WMT), pages 789–799, Suzhou, China. Association for Computational Linguistics… find
  11. Tom Kocmi, Eleftherios Avramidis, Rachel Bawden, Ondřej Bojar, and 1 others. 2025b. Findings of the WMT25 general machine translation shared task. In Proceedings of the Tenth Conference on Machine Translation (WMT). Association for Computational Linguistics. find
  12. H. Li, J. Chen, and Q. Liu. 2025. MAATS: A multi-agent automated translation system based on MQM evaluation. arXiv preprint arXiv:2505.14848. find
  13. Yasmin Moslem, Rejwanul Haque, John D. Kelleher, and Andy Way. 2023. Adaptive machine translation with large language models. In Proceedings of the 24th Annual Conference of the European Association for Machine Translation, pages 227–237.
  14. Maja Popović. 2015. chrF: Character n-gram F-score for automatic MT evaluation. In Proceedings of the Tenth Workshop on Statistical Machine Translation, pages 392–395. Association for Computational Linguistics.
  15. Kirill Semenov and Ondřej Bojar. 2022. Automated evaluation metric for terminology consistency in MT. In Proceedings of the Seventh Conference on Machine Translation (WMT), pages 450–457. Association for Computational Linguistics. find
  16. Kirill Semenov, Xu Huang, Vilém Zouhar, Nathaniel Berger, Dawei Zhu, Arturo Oncevay, and Pinzhen Chen. 2025. Findings of the WMT25 terminology translation task: Terminology is useful especially for good MTs. In Proceedings of the Tenth Conference on Machine Translation (WMT), pages 554–576, Suzhou,… find
  17. Kirill Semenov, Vilém Zouhar, Tom Kocmi, Dongdong Zhang, Wangchunshu Zhou, and Yuchen Eleanor Jiang. 2023. Findings of the WMT 2023 shared task on machine translation with terminologies. In Proceedings of the Eighth Conference on Machine Translation (WMT), pages 663–671, Singapore. Association for C… find
  18. Minghao Wu, Yuan He, Yufei Zhang, Xiaojun Xu, Maosong Sun, Longyue Huang, and Lemao Liu. 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 245–256, Miami, Flo… find
  19. Jiawei Zhang, Yuzhi Tian, Junjie Yin, Min Yang, and Bing Qin. 2024a. TACTIC: Translation agents with cognitive-theoretic interactive collaboration. arXiv preprint arXiv:2506.08403. — [arXiv:2506.08403](https://arxiv.org/abs/2506.08403) find
  20. Liang Zhang, Siyuan Peng, and Hongbin Liu. 2024b. GRAFT: A graph-based flow-aware agentic framework for document-level machine translation. arXiv preprint arXiv:2507.03311. *Figure 3: ChrF2++ vs. accuracy for en→de (Proper mode).* Yang Zhao, Jiajun Liu, and Hao Wang. 2025. Are AI agents the new mach… find
  21. Lichao Zhu, Maria Zimina-Poirot, Stephane Patin, and Cristian Valdez. 2025. Laniqo at WMT25 terminology translation task: A multi-objective reranking strategy for terminology-aware translation via Pareto-optimal decoding. In Proceedings of the Tenth Conference on Machine Translation (WMT), Suzhou, C… find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/rosa-2026-multi-agent-orchestration-terminology-constrained/