Elamur Research⌕
☰
home / papers / 2026 / Parallel Corpora of Scholarly Documents for English-French Machine Translation

Parallel Corpora of Scholarly Documents for English-French Machine Translation

Summary

LLMs can now handle long inputs, which makes document-level machine translation of scholarly writing feasible. However, English–French parallel data at the document level is rare for scientific texts. Most existing resources are sentence-aligned, limited to medicine, or contain only abstracts. This leaves little material for adapting or properly evaluating systems on whole articles. The authors build two corpora to address this: PARAEPS (Earth and Planetary Sciences) and PARANLP (Natural Language Processing). Each has train, dev and test sets of abstracts, plus a TEST-LONG set of complete articles. The article references come from three sources: human translations, professional post-edits of MT output, and a new pipeline that turns comparable English and French NLP publications into a partly machine-generated "silver" reference. The authors benchmark eight MT systems. Fine-tuning medium-size open LLMs on the new data brings them to roughly the level of commercial tools such as DeepL Pro on abstracts. Translating entire articles in one pass gives no consistent gain over paragraph-level translation and sometimes performs worse. The resources and code are openly released, supporting work on discourse-aware scientific MT.

Key points

  • Size: about 14k parallel abstracts and 105 full-length articles in total.
    • PARAEPS: 10,577 / 400 / 391 abstracts (train/dev/test) and 29 articles (MERSENNE 19, STUDENT 10).
    • PARANLP: 2,723 / 96 / 346 abstracts and 76 articles (NLPGOLD 4, NLPSILVER EN-FR 36, NLPSILVER FR-EN 36).
  • Processing pipeline:
    • langdetect for language filtering.
    • Trankit for sentence segmentation.
    • A modified BertAlign for alignment, with skip set to 0.001 and a new len_slack parameter set to 0.15.
    • TransQuest quality-estimation scores for filtering. In MERSENNE, all aligned sentence pairs scored at least 0.75.
  • Sources of full articles:
    • STUDENT: translations made by master's students in a translation course, using ISTEX or Open Access articles.
    • MERSENNE: articles from "Comptes rendus Géosciences", machine-translated and then post-edited in MateCat by a professional translator.
  • Results: fine-tuning narrows the gap between open LLMs and commercial MT. Article-level translation with recent LLMs offers no systematic benefit over paragraph-level translation.
  • Availability: the corpora and pipeline code are released under a permissive license through the ANR MaTOS project.

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: Linguistics & languages · Machine translation · Natural language processing

Topics: Machine translation

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/2jm9pkbjkg95

@inproceedings{peng2026parallel,
  title = {{Parallel Corpora of Scholarly Documents for English-French Machine Translation}},
  author = {Peng, Ziqian and Zhu, Lichao and Bawden, Rachel and Bénard, Maud and Clergerie, Éric de la and Huguin, Mathilde and Kübler, Natalie and Lerner, Paul and Mestivier, Alexandra and Yvon, François},
  booktitle = {Proceedings of the 19th Workshop on Building and Using Comparable Corpora (BUCC)},
  year = {2026},
  doi = {10.63317/2jm9pkbjkg95},
  url = {https://doi.org/10.63317/2jm9pkbjkg95}
}

References

18 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Dániel Varga, Péter Halaácsy, András Kornai, Vok-tor Nagy, László Németh, and Viktor Trón. 2005. Parallel corpora for medium density languages. In Proceedings of the RANLP 2005 Conference, pages 590–596. find
  2. Esalati, Mersad and Dousti, Mohammad Javad and Faili, Heshaam. 2024. Esposito: An English-Persian Scientific Parallel Corpus for Machine Translation. Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024). ELRA and ICCL. PID https://huggingface.co/datasets/universitytehran. ↗
  3. Giorgos Vernikos, Brian Thompson, Prashant Mathur, and Marcello Federico. 2022. Embarrassingly easy document-level MT metrics: How to convert any pretrained metric into a document-level metric. In Proceedings of the Seventh Conference on Machine Translation (WMT), pages 118–128, Abu Dhabi, United Arab Emirates (Hybrid). Association for Computational Linguistics. find our card
  4. Ive, Julia and Max, Aurélien and Yvon, François and Ravaud, Philippe. 2016. Diagnosing High-Quality Statistical Machine Translation Using Traces of Post-Edition Operations. International Conference on Language Resources and Evaluation - Workshop on Translation Evaluation: From Fragmented Tools and Data Sets to an Integrated Ecosystem (MT Eval 2016). find
  5. Longyue Wang, Zefeng Du, Wenxiang Jiao, Chenyang Lyu, Jianhui Pang, Leyang Cui, Kaiqiang Song, Derek Wong, Shuming Shi, and Zhaopeng Tu. 2024. Benchmarking and improving long-text translation with large language models. In Findings of the Association for Computational Linguistics: ACL 2024, pages 7175– 7187, Bangkok, Thailand. Association for Computational Linguistics. find
  6. Hannah Calzi Kleidermacher and James Zou. 2025. Science Across Languages: Assessing LLM Multilingual Translation of Scientific Papers. PID https://arxiv.org/abs/2502.17882. ↗
  7. Iñaki Lacunza and Javier Garcia Gilabert and Francesca De Luca Fornaciari and Javier Aula-Blasco and Aitor Gonzalez-Agirre and Maite Melero and Marta Villegas. 2025. ACADATA: Parallel Dataset of Academic Data for Machine Translation. https://huggingface.co/datasets/BSC-PID LT/ACAData. ↗
  8. Yutong Wang, Jiali Zeng, Xuebo Liu, Derek F. Wong, Fandong Meng, Jie Zhou, and Min Zhang. 2025. DelTA: An online document-level translation agent based on multi-level memory. In The Thirteenth International Conference on Learning Representations. ↗
  9. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, Chujie Zheng, Dayiheng Liu, Fan Zhou, Fei Huang, Feng Hu, Hao Ge, Haoran Wei, Huan Lin, Jialong Tang, Jian Yang, Jianhong Tu, Jianwei Zhang, Jianxin Yang, and Jiaxi Yang et al. Qwen3 technical report. Preprint 2025. arXiv:2505.09388. ↗ our card
  10. Denis Maurel, Enza Morale, Nicolas Thouvenin, Patrice Ringot, and Angel Turri. 2019. Istex: A database of twenty million scientific papers with a mining tool which uses named entities. Information, 10(5). ↗
  11. Névéol, Aurélie and Jimeno Yepes, Antonio and Neves, Mariana and Verspoor, Karin. Parallel Corpora for the Biomedical 2018. find
  12. Domain. Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018). European Language Resources Association (ELRA). PID https://github.com/biomedical-translation-corpora/corpora. ↗
  13. Dayyán O’Brien and Bhavitvya Malik and Ona de Gibert and Pinzhen Chen and Barry Haddow and Jörg Tiedemann. 2025. DocHPLT: A Massively Multilingual PID Document-Level Translation Dataset. https://huggingface.co/datasets/HPLT/DocHPLT. ↗
  14. Roussis, Dimitrios and Papavassiliou, Vassilis and Prokopidis, Prokopis and Piperidis, Stelios and Katsouros, Vassilis. 2022. SciPar: A Collection of Parallel Corpora from Scientific Abstracts. Proceedings of the Thirteenth Language Resources and Evaluation Conference. European Language Resources Association. PID https://live.european-language-grid.eu/catalogue/corpus/20067. ↗
  15. Salesky, Elizabeth and Darwish, Kareem and Al-Badrashiny, Mohamed and Diab, Mona and Niehues, Jan. 2023. Evaluating Multilingual Speech Translation under Realistic Conditions with Resegmentation and Terminology. Association for Computational Linguistics. PID https://iwslt.org/2023/multilingual. ↗
  16. Wang, Longyue and Du, Zefeng and Jiao, Wenxiang and Lyu, Chenyang and Pang, Jianhui and Cui, Leyang and Song, Kaiqiang and Wong, Derek and Shi, Shuming and Tu, Zhaopeng. 2024a. Benchmarking and Improving Long-Text Translation with Large Language Models. Findings of the Association for Computational Linguistics: ACL 2024. Association for Computational Linguistics. PID https://github.com/longyuewangdcu/Document-MT-LLM. ↗
  17. Wang, Longyue and Liu, Siyou and Lyu, Chenyang and Jiao, Wenxiang and Wang, Xing and Xu, Jiahao and Tu, Zhaopeng and Gu, Yan and Chen, Weiyu and Wu, Minghao and Zhou, Liting and Koehn, Philipp and Way, Andy Findings of the and Yuan, Yulin. 2024b. find
  18. WMT 2024 Shared Task on Discourse-Level Proceedings of the Literary Translation. Ninth Conference on Machine Translation. Association for Computational Linguistics. https://www2.statmt.org/wmt24/literary-PID translation-task.html. ↗

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/peng-2026-parallel-corpora-scholarly-documents-english/