Elamur Research⌕
☰
home / papers / 2026 / BanSuite: A Unified Toolkit and Software Platform for Low-Resource NLP in Bangla

BanSuite: A Unified Toolkit and Software Platform for Low-Resource NLP in Bangla

Paper · 2026 · Md. Abu Sayed, Faisal Ahamed Khan, Jannatul Ferdous Tuli et al.

Summary

Bangla has a very large speaker base but lacks an integrated NLP stack comparable to spaCy or Stanford CoreNLP. Earlier Bangla toolkits (BNLP, iNLTK) mostly handle preprocessing, embeddings or tagging and provide no syntactic parsers. BanSuite, built under the EBLICT project, fills this gap. It rests on an in-house, linguist-validated Bangla treebank of about 170K sentences covering news, social media, literature and dialogue. The treebank carries four annotation layers: POS, NER, shallow parsing (SP) and dependency parsing (DP). Each sentence was labeled by three annotators, a majority vote set the initial label, and an expert validator made the final decision. The trained models are distributed through Bkit, a Python library that also offers tokenization, normalization, text cleaning and lemmatization, and through a web application for non-programmers. In-domain test scores are around 87–90. In zero-shot benchmarks, BanSuite beats BNLP and several LLMs on every task while using far less power. This makes it a practical, open foundation for Bangla research and applications.

Key points

  • Treebank: about 170–174K sentences and roughly 2.0M words per task. Inter-annotator agreement (Fleiss' κ) is 0.92 for NER, 0.77 for POS, and 0.68 for both SP and DP. Validation acceptance rates range from 90.5% to 98.2%.
  • In-domain results: POS 90.16 F1, NER 90.11 F1, SP 86.92 F1 and DP 90.27 UAS.
  • Zero-shot comparison: BanSuite reaches 88.78 F1 on NER, compared with 56.29 for Gemini 2.5F. Its other zero-shot scores are an average POS F1 of 79.93, an SP F1 of 52.56, and a DP UAS of 72.19. Baselines include BNLP, Llama 4 Scout, Gemma 3 27B and GPT-OSS-20B. Few-shot prompting improved Gemma 3 by about 11.37%.
  • Efficiency: BanSuite (T5) achieves about 72% task performance at under 100 W. The LLMs score about 41–51% and draw over 200 W. BNLP (CRF) uses about 30 W but reaches only about 27%.
  • Limitations: the system covers Bangla only, may perform worse on informal or code-mixed text, and relies on tag mapping to compare against BNLP.

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

🎓Fordham UniversityNew York · United States4 works · 737 citations

Also on the byline: North South University.

 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.Low-resource languages · 267 works

Cite

Cite as Md. Abu Sayed, Faisal Ahamed Khan, Jannatul Ferdous Tuli, Nabeel Mohammed, Mohammad Ruhul Amin and Mohammad Mamun Or Rashid. 2026. BanSuite: A Unified Toolkit and Software Platform for Low-Resource NLP in Bangla. Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 3: System Demonstrations).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.eacl-demo.44

@inproceedings{sayed2026bansuite,
  title = {{BanSuite: A Unified Toolkit and Software Platform for Low-Resource NLP in Bangla}},
  author = {Md. Abu Sayed, Faisal Ahamed Khan, Jannatul Ferdous Tuli, Nabeel Mohammed, Mohammad Ruhul Amin and Mohammad Mamun Or Rashid},
  booktitle = {Proceedings of the 19th Conference of the European Chapter of the Association for Computational Linguistics (Volume 3: System Demonstrations)},
  year = {2026},
  doi = {10.18653/v1/2026.eacl-demo.44},
  url = {https://doi.org/10.18653/v1/2026.eacl-demo.44}
}

References

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

  1. Sadia Afrin, Md. Shahad Mahmud Chowdhury, Md. Islam, Faisal Khan, Labib Chowdhury, Md. Mahtab, Nazifa Chowdhury, Massud Forkan, Neelima Kundu, Hakim Arif, Mohammad Mamun Or Rashid, Mohammad Amin, and Nabeel Mohammed. 2023. Ban-Lemma: A word formation dependent rule and dictionary based Bangla lemmat… find
  2. Gaurav Arora. 2020. iNLTK: Natural language toolkit for indic languages. In Proceedings of Second Workshop for NLP Open Source Software (NLP-OSS), pages 66–71, Online. Association for Computational Linguistics. find
  3. Kalika Bali, Monojit Choudhury, and Priyanka Biswas. 2010. Indian language part-of-speech tagset: Bengali (ldc2010t16). Linguistic Data Consortium, Philadelphia. https://catalog.ldc.upenn.edu/ LDC2010T16. Web download. http://eblict.gov.bd/ 7 https://bcc.gov.bd/ 8 https://ictd.gov.bd/ 9 https://giga… ↗
  4. Steven Bird, Edward Loper, and Ewan Klein. 2009. Natural Language Processing with Python. O’Reilly Media Inc. find
  5. IIT Kharagpur. 2005. Indian language pos-tagged corpus. https://www.nltk.org/api/nltk.corpus. reader.indian.html. Accessed: 2025-11-11. ↗
  6. George Cardona and Dhanesh Jain. 2007. The Indo-Aryan Languages. Routledge. find
  7. Nikita Kitaev, Steven Cao, and Dan Klein. 2019. Multilingual constituency parsing with self-attention and pre-training. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 3499–3505, Florence, Italy. Association for Computational Linguistics.
  8. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. Preprint, arXiv:1810.04805.
  9. David M Eberhard, Gary F Simons, and Charles D Fennig. 2024. Ethnologue: Languages of the World, 27th edition. SIL International, Dallas, Texas. find
  10. Joseph L. Fleiss. 1971. Measuring nominal scale agreement among many raters. Psychological Bulletin, 76(5):378–382. find
  11. Matt Gardner, Joel Grus, Mark Neumann, Oyvind Tafjord, Pradeep Dasigi, Nelson F. Liu, Matthew Peters, Michael Schmitz, and Luke S. Zettlemoyer. 2018. Allennlp: A deep semantic natural language processing platform. arXiv preprint arXiv:1803.07640.
  12. Google DeepMind. 2025. Gemini 2.5 flash. https://ai.google.dev/gemini-api/. Accessed: 2025-11-21. ↗
  13. Md. Zahidul Haque, Sakib Zaman, Jillur Rahman Saurav, Summit Haque, Md. Saiful Islam, and Mohammad Ruhul Amin. 2023. B-ner: A novel bangla named entity recognition dataset with largest entities and its baseline evaluation. IEEE Access, 11:45194– 45205. find
  14. Nikita Kitaev and Dan Klein. 2018. Constituency parsing with a self-attentive encoder. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 2676–2686, Melbourne, Australia. Association for Computational Linguistics.
  15. John D. Lafferty, Andrew McCallum, and Fernando C. N. Pereira. 2001. Conditional random fields: Probabilistic models for segmenting and labeling sequence data. In Proceedings of the Eighteenth International Conference on Machine Learning, ICML ’01, page 282–289, San Francisco, CA, USA. Morgan Kaufma… find
  16. Boda Lin, Zijun Yao, Jiaxin Shi, Shulin Cao, Binghao Tang, Si Li, Yong Luo, Juanzi Li, and Lei Hou. 2022. Dependency parsing via sequence generation. In Findings of the Association for Computational Linguistics: EMNLP 2022, pages 7339–7353, Abu Dhabi, United Arab Emirates. Association for Computatio… find
  17. Matthew Honnibal, Ines Montani, Sofie Van Lan-deghem, and Adriane Boyd. 2020. spacy: Industrial-strength natural language processing in python. https://spacy.io. ↗
  18. Kun Liu, Yao Fu, Chuanqi Tan, Mosha Chen, Ningyu Zhang, Songfang Huang, and Sheng Gao. 2021. Noisy-labeled NER with confidence estimation. In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 3437–344… find
  19. Md. Ekramul Islam, Labib Chowdhury, Faisal Ahamed Khan, Shazzad Hossain, Md Sourave Hossain, Mohammad Mamun Or Rashid, Nabeel Mohammed, and Mohammad Ruhul Amin. 2023. Sentigold: A large bangla gold standard multi-domain sentiment analysis dataset and its evaluation. In Proceedings of the 29th ACM SI… find
  20. Md. Motahar Mahtab, Faisal Ahamed Khan, Md. Ekramul Islam, Md. Shahad Mahmud Chowdhury, Labib Imam Chowdhury, Sadia Afrin, Hazrat Ali, Mohammad Mamun Or Rashid, Nabeel Mohammed, and Mohammad Ruhul Amin. 2025. BanNERD: A benchmark dataset and context-driven approach for Bangla named entity recognitio… find
  21. Robert A. Jacobs, Michael I. Jordan, Steven J. Nowlan, and Geoffrey E. Hinton. 1991. Adaptive mixtures of local experts. Neural Computation, 3(1):79–87. find
  22. Christopher D. Manning, Mihai Surdeanu, John Bauer, Jenny Finkel, Steven J. Bethard, and David Mc-Closky. 2014. The Stanford CoreNLP natural language processing toolkit. In Association for Computational Linguistics (ACL) System Demonstrations, pages 55–60.
  23. Siratun Jannat, Mizanur Rahoman, Shafi Sourov, Jannatul Ferdaousi, Syeda Shahzadi, and Daniel Zeman. 2021. Ud bengali bru (universal dependencies v2.9). https://github.com/UniversalDependencies/ UD_BengaliâĂŚBRU. Accessed: 2025-11-19; License: CCBY-SA4.0. ↗
  24. Meta AI. 2025. Llama 4 scout: A natively multimodal mixture-of-experts model. Version 17B-16E, released April 5, 2025. find
  25. Arnav Mhaske, Harshit Kedia, Sumanth Doddapaneni, Mitesh M. Khapra, Pratyush Kumar, Rudra Murthy, and Anoop Kunchukuttan. 2023. Naamapadam: A large-scale named entity annotated data for Indic languages. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volum… find
  26. Pruthwik Mishra, Vandan Mujadia, and Dipti Misra Sharma. 2024. Multi task learning based shallow parsing for indian languages. ACM Trans. Asian Low-Resour. Lang. Inf. Process., 23(9). find
  27. OpenAI, :, Sandhini Agarwal, Lama Ahmad, Jason Ai, Sam Altman, Andy Applebaum, Edwin Arbus, Rahul K. Arora, Yu Bai, Bowen Baker, Haiming Bao, Boaz Barak, Ally Bennett, Tyler Bertao, Nivedita Brett, Eugene Brevdo, Greg Brockman, Sebastien Bubeck, and 108 others. 2025. gpt-oss-120b gpt-oss-20b model c…

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/sayed-2026-bansuite-unified-toolkit-software-platform/