Elamur Research⌕
☰
home / papers / 2026 / Introducing a Bangla Sentence – Gloss Pair Dataset for Bangla Sign Language Translation and Research

Introducing a Bangla Sentence – Gloss Pair Dataset for Bangla Sign Language Translation and Research

Summary

Research on Bangla Sign Language (BdSL) has mostly addressed recognition of isolated letters and words, because there are almost no sentence-level resources. The authors address this with Bangla-SGP, a parallel corpus that pairs Bangla sentences with BdSL gloss sequences. A gloss is a textual stand-in for a sign, and a sequence of glosses can serve as an intermediate step toward sign video or 3D avatar generation.

The core of the corpus is 1,000 sentences drawn from newspapers, literature and BdSL teaching material, each glossed by a professional signer. The team then added roughly 3,000 synthetic pairs using three methods:

  • tense rules derived together with the signer,
  • masked-token substitution with Bangla-Bert,
  • a RAG-based generation step guided by rules and prompts.

They fine-tuned mBart50, Google mT5 and GPT4.1-nano for sentence-to-gloss translation, scored them with BLEU, and compared their behaviour on Bangla-SGP with RWTH-PHOENIX-2014T. The supplied excerpt omits the results section, so exact scores are not given here.

The resource matters because Bangladesh has millions of deaf and hard-of-hearing people with very little language technology in BdSL.

Key points

  • Corpus size: Bangla-SGP holds about 4,000 sentence–gloss pairs, 1,000 human-annotated and about 3,000 augmented, a 1:3 ratio. The annotator was Ariful Islam, a BdSL presenter at BTV.
  • Tense-rule augmentation: Rules mapping verbs to gloss form were applied to rewrite sentences into other tenses, producing about 500 pairs. In future tense, for example, the verb root is followed by "hobe".
  • Models and evaluation: mBart50, mT5 and GPT4.1-nano were evaluated with BLEU against the RWTH-PHOENIX-2014T benchmark. NLLB-200-1.3B was not tested on the augmented data because of limited compute.
  • Phase 1 scope: The released corpus contains gloss text only. Phase 2 plans synchronized sign videos, non-manual marker annotation by several signers, and inter-annotator agreement checks.
  • Planned 3D pipeline: A gloss-to-video dictionary would feed SMPLer-X to produce 3D sign animation. Early attempts ran into unstable transitions between consecutive signs.

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 10.63317/38qenrwzegr9

@inproceedings{saha2026introducing,
  title = {{Introducing a Bangla Sentence – Gloss Pair Dataset for Bangla Sign Language Translation and Research}},
  author = {Saha, Neelavro and Shahriyar, Rafi and Roudra, Nafis Ashraf and Sakib, Saadman and Rasel, Annajiat Alim},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation Conference},
  year = {2026},
  doi = {10.63317/38qenrwzegr9},
  url = {https://doi.org/10.63317/38qenrwzegr9}
}

References

16 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Mohammad Alauddin and Abul Hasnat Joarder. 2004. Deafness in Bangladesh, pages 64–69. Springer Japan, Tokyo. find
  2. Zhongang Cai, Wanqi Yin, Ailing Zeng, Chen Wei, Qingping Sun, Yanjun Wang, Hui En Pang, Haiyi Mei, Mingyuan Zhang, Lei Zhang, Chen Change Loy, Lei Yang, and Ziwei Liu. 2024. Smpler-x: Scaling up expressive human pose and shape estimation. find
  3. Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2019. Unsupervised cross-lingual representation learning at scale. CoRR, abs/1911.02116. find
  4. Marta R Costa-Jussà, James Cross, Onur Çelebi, Maha Elbayad, Kenneth Heafield, Kevin Heffernan, Elahe Kalbassi, Janice Lam, Daniel Licht, Jean Maillard, et al. 2022. No language left behind: Scaling human-centered machine translation. arXiv preprint arXiv:2207.04672. find
  5. Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lora: Low-rank and Weizhu Chen. 2021. adaptation of large language models. CoRR, abs/2106.09685. find
  6. Hamid Reza Vaezi Joze and Oscar Koller. 2019. Ms-asl: A large-scale data set and benchmark for understanding american sign language. find
  7. 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
  8. Amit Moryossef, Mathias Müller, Anne Göhring, Zifan Jiang, Yoav Goldberg, and Sarah Ebling. 2023. An open-source gloss-based baseline for spoken to signed language translation. find
  9. Georgios Pavlakos, Vasileios Choutas, Nima Ghorbani, Timo Bolkart, Ahmed A. Osman, Dimitrios Tzionas, and Michael J. Black. 2019. Expressive Body Capture: 3D Hands, Face, and Body From a Single Image . In 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 10967–10977,… find
  10. Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter Liu. 2019. Exploring the limits of transfer learning with a unified text-to-text transformer. find
  11. Husne Rubaiyeat, Hasan Mahmud, Ahsan Habib, and Md Kamrul Hasan. 2024. Bdslw60: A word-level bangla sign language dataset. find
  12. Ataher Sams, Ahsan Habib Akash, and S. M. Mahbubur Rahman. 2023. Signbd-word: Video-based bangla word-level sign language and pose translation. In 2023 14th International Conference on Computing Communication and Networking Technologies (ICCCNT), pages 1–7. find
  13. 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. CoRR, abs/2008.00401. find
  14. Beverly Trezek and Connie Mayer. 2019. Reading and deafness: State of the evidence and implications for research and practice. Education Sciences, 9(3). find
  15. 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. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Ling… find
  16. Ronglai Zuo, Fangyun Wei, Zenggui Chen, Brian Mak, Jiaolong Yang, and Xin Tong. 2024. A simple baseline for spoken language to sign language translation with 3d avatars. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/papers/saha-2026-introducing-bangla-sentence-gloss-pair/