Elamur Research⌕
☰
home / papers / 2026 / FormosanMT: A Multilingual Parallel Corpus of the Formosan Language Family

FormosanMT: A Multilingual Parallel Corpus of the Formosan Language Family

Summary

Machine translation for endangered languages is held back by a shortage of usable parallel data, which in turn slows the development of MT methods for extremely low-resource settings. This paper addresses that gap for the indigenous languages of Taiwan by releasing FormosanMT. It contains about half a million sentence pairs linking 15 Formosan languages (plus Yami) with Mandarin and, where data allows, English. The data comes from public material curated in FormosanBank and from ILRDF video subtitles. The authors also publish reproducible pipelines that extract, clean, normalize and split the data, so the corpora can be rebuilt or grown as the source repositories expand.

Beyond the data, they add a Traditional Chinese tokenizer for NLLB, which previously handled only Simplified Chinese. They also provide updated NLLB fine-tuning code that works for any low-resource language pair. Using NLLB and OpenNMT, they train baseline translation models for every language with Mandarin and for four languages with English.

Because the corpus covers many related languages, it can serve as a testbed for cross-lingual transfer in endangered-language MT. The resources are also useful to the Formosan speaker communities themselves.

Key points

  • The corpus has roughly 500,000 Formosan–Mandarin and Formosan–English sentence pairs. It covers 15 languages, with Seediq and Truku merged under one ISO code (trv) and Yami/Tao included.
  • The main sources are FormosanBank collections: the ILRDF Dictionary, ePark, the NTU Corpus and the Presidential Apology. Scripts are provided to scrape ILRDF video subtitles, which are not redistributed until permission is confirmed.
  • Bible translations were deliberately excluded. The reason is that their domain and style mismatch can increase hallucinations in MT output.
  • The new tools are a Traditional Chinese tokenizer compatible with NLLB, refreshed NLLB fine-tuning code, and extraction scripts such as fetch_xml.py, clean_xml.py, make_corpus.py and scrape_videos.py. The extraction keeps dialect labels for each pair.
  • Trained NLLB and OpenNMT baselines are released for all 15 languages with Mandarin and for 4 languages with English. Everything is available on GitHub (FormosanBank/FormosanBank-MT).

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

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/3ut33xim33nt

@inproceedings{scheppat2026formosanmt,
  title = {{FormosanMT: A Multilingual Parallel Corpus of the Formosan Language Family}},
  author = {Scheppat, Hunter and Hartshorne, Joshua K. and Koc, Sema and Ferrand, Éric Le and Prud'hommeaux, Emily},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation Conference},
  year = {2026},
  doi = {10.63317/3ut33xim33nt},
  url = {https://doi.org/10.63317/3ut33xim33nt}
}

References

40 references, 8 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Aboriginal Language Research and Development Foundation. 2023a. Online dictionary of aboriginal languages. https://e-dictionary.ilrdf.org.tw. ↗
  2. Aboriginal Language Research and Development Foundation. 2023b. Yuanzhumin yuyan leyuan (epark). https://web.klokah.tw/. ↗
  3. Željko Agić and Ivan Vulić. 2019. Jw300: A wide-coverage parallel corpus for low-resource languages. In 7th Workshop on Balto-Slavic Natural Language Processing. Association for Computational Linguistics, pages 3204–3210. find our card
  4. Peter Bellwood. 1984. A hypothesis for austronesian origins. Asian Perspectives, 26(1):107–117. find
  5. Steven Bird and David Chiang. 2012. Machine translation for language preservation. In Proceedings of COLING 2012: Posters, pages 125– 134. find
  6. Robert Blust. 2019. The austronesian homeland and dispersal. Annual Review of Linguistics, 5(1):417–434. find
  7. Marta R. Costa-jussà, James Cross, Onur Çelebi, Maha Elbayad, Kenneth Heafield, Kevin Heffernan, Elahe Kalbassi, Janice Lam, Daniel Licht, Jean Maillard, Anna Sun, Skyler Wang, Guillaume Wenzek, Al Youngblood, Bapi Akula, Loic Barrault, Gabriel Mejia Gonzalez, Prangthip Hansanti, John Hoffman, Semar… find
  8. Pedro Henrique Domingues, Claudio Santos Pinhanez, Paulo Cavalin, and Julio Nogima. 2024. Quantifying the ethical dilemma of using culturally toxic training data in ai tools for indigenous languages. In Proceedings of the 3rd Annual Meeting of the Special Interest Group on Under-resourced Languages@… find
  9. David M Eberhard, Gary F Simons, and Charles D Fennig. 2024. Ethnologue: languages of the world, 27th Edition, volume 22. SIL International. find
  10. Virginia Fey. 1986. Amis dictionary. Taipei: The Bible Society. find
  11. Barry Haddow, Rachel Bawden, Antonio Valerio Miceli Barone, Jindřich Helcl, and Alexandra Birch. 2022. Survey of low-resource machine translation. Computational Linguistics, 48(3):673–732. find our card
  12. Joshua K. Hartshorne, Éric Le Ferrand, Li-May Sung, and Emily Prud’hommeaux. 2024. Formosanbank and why you should use it. In Architectures and Mechanisms in Language Processing (AMLaP) Poster. find
  13. Marcin Junczys-Dowmunt, Roman Grundkiewicz, Tomasz Dwojak, Hieu Hoang, Kenneth Heafield, Tom Neckermann, Frank Seide, Ulrich Germann, Alham Fikri Aji, Nikolay Bogoychev, André F. T. Martins, and Alexandra Birch. 2018. Marian: Fast neural machine translation in C++. In Proceedings of ACL 2018, System… find our card
  14. Katharina Kann, Abteen Ebrahimi, Kristine Stenzel, and Alexis Palmer. 2022. Machine translation between high-resource languages in a language documentation setting. In Proceedings of 1st Workshop on NLP applications to field linguistics, page 26. find
  15. Guillaume Klein, François Hernandez, Vincent Nguyen, and Jean Senellart. 2020. The Open-NMT neural machine translation toolkit: 2020 edition. In Proceedings of the 14th Conference of the Association for Machine Translation in the Americas (Volume 1: Research Track), pages 102–109. find
  16. Taku Kudo and John Richardson. 2018. Sentencepiece: A simple and language independent subword tokenizer and detokenizer for neural text processing. find our card
  17. Yinhan Liu, Jiatao Gu, Naman Goyal, Xian Li, Sergey Edunov, Marjan Ghazvininejad, Mike Lewis, and Luke Zettlemoyer. 2020. Multilingual denoising pre-training for neural machine translation. Transactions of the Association for Computational Linguistics, 8:726–742. find
  18. Gerardo Sierra Martínez, Cynthia Montaño, Gemma Bel-Enguix, Diego Córdova, and Margarita Mota Montoya. 2020. Cplm, a parallel corpus for mexican languages: Development and interface. In Proceedings of the Twelfth Language Resources and Evaluation Conference, pages 2947–2952. find
  19. Thomas Mayer and Michael Cysouw. 2014. Creating a massively parallel Bible corpus. In Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC 14), pages 3158–3163. find our card
  20. So Miyagawa. 2023. Machine translation for highly low-resource language: A case study of ainu, a critically endangered indigenous language in northern japan. In Proceedings of the Joint 3rd International Conference on Natural Language Processing for Digital Humanities and 8th International Workshop… find
  21. Wael Mohamed, Éric Le Ferrand, Li-May Sung, Emily Prud’hommeaux, and Joshua Hartshorne. 2024. Formosanbank. https://ai4commsci.gitbook.io/formosanbank. ↗
  22. Anna Mossolova and Kamel Smaïli. 2022. The only chance to understand: machine translation of the severely endangered low-resource languages of eurasia. In The Fifth Workshop on Technologies for Machine Translation of Low-Resource Languages (LoResMT), COLING 2022. find
  23. Atul Kr. Ojha, Chao-hong Liu, Ekaterina Vylomova, Flammie Pirinen, Jonathan Washington, Nathaniel Oco, and Xiaobing Zhao, editors. 2025. Proceedings of the Eighth Workshop on Technologies for Machine Translation of Low-Resource Languages (LoResMT 2025). find
  24. Arturo Oncevay. 2021. Peru is multilingual, its machine translation should be too? In Proceedings of the First Workshop on Natural Language Processing for Indigenous Languages of the Americas, pages 194–201. find
  25. John E Ortega, Richard Castro Mamani, and Kyunghyun Cho. 2020. Neural machine translation with a polysynthetic low resource language. Machine Translation, 34(4):325–346. find
  26. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. Bleu: a method for automatic evaluation of machine translation. In Proceedings of the 40th annual meeting of the Association for Computational Linguistics, pages 311–318. find our card
  27. Maja Popović. 2017. chrf++: words helping character n-grams. In Proceedings of the Second Conference on Machine Translation, pages 612– 618. find our card
  28. Matt Post. 2018. A call for clarity in reporting {BLEU} scores. In Proceedings of the Third Conference on Machine Translation, page 186–191. find our card
  29. Juan Prieto, Cristian Martinez, Melissa Robles, Alberto Moreno, Sara Palacios, and Rubén Manrique. 2024. Translation systems for low-resource colombian indigenous languages, a first step towards cultural preservation. In Proceedings of the 4th Workshop on Natural Language Processing for Indigenous L… find
  30. Hunter Scheppat, Joshua Hartshorne, Dylan Leddy, Eric Le Ferrand, and Emily Prudhommeaux. 2025. Integrating diverse corpora for training an endangered language machine translation system. In Proceedings of the Eight Workshop on the Use of Computational Methods in the Study of Endangered Languages, p… find
  31. Jiatong Shi, Jonathan D Amith, Xuankai Chang, Siddharth Dalmia, Brian Yan, and Shinji Watanabe. 2021. Highland puebla nahuatl speech translation corpus for endangered language documentation. In Proceedings of the First Workshop on Natural Language Processing for Indigenous Languages of the Americas,… find
  32. Lily I Su, Li-May Sung, Shuping Huang, Fuhui Hsieh, and Zhemin Lin. 2008. Ntu corpus of formosan languages: A state-of-the-art report. Corpus Linguistics & Linguistic Theory, 4(2). find
  33. Li-May Sung, I Lily, Fuhui Hsieh, and Zhemin Lin. 2008. Developing an online corpus of formosan languages. Taiwan Journal of Linguistics, 6(2). find
  34. Yuqing Tang, Chau Tran, Xian Li, Peng-Jen Chen, Naman Goyal, Vishrav Chaudhary, Jiatao Gu, and Angela Fan. 2020. Multilingual translation with extensible multilingual pretraining and fine-tuning. arXiv preprint arXiv:2008.00401. find
  35. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. Attention is all you need. Advances in neural information processing systems, 30. find
  36. Haifeng Wang, Hua Wu, Zhongjun He, Liang Huang, and Kenneth Ward Church. 2022. Progress in machine translation. Engineering, 18:143–153. find
  37. Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, and Colin Raffel. 2021. mt5: A massively multilingual pre-trained text-to-text transformer. find
  38. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 483–498. find
  39. Shiyue Zhang, Benjamin Frey, and Mohit Bansal. 2020. Chren: Cherokee-english machine translation for endangered language revitalization. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 577–595. find
  40. Francis Zheng, Edison Marrese-Taylor, and Yutaka Matsuo. 2022. A parallel corpus and dictionary for amis-mandarin translation. In Proceedings of the 2nd International Workshop on Natural Language Processing for Digital Humanities, pages 79–84. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/scheppat-2026-formosanmt-multilingual-parallel-corpus-formosan/