Elamur Research⌕
☰
home / papers / 2026 / POS Tagging in Low-Resource Maithili Language: Specific Challenges and Nuances

POS Tagging in Low-Resource Maithili Language: Specific Challenges and Nuances

Paper · 2026 · Shivani Priya, Shruti Jha, Urmila Jha et al.

Summary

Maithili is an Indo-Aryan language spoken in Bihar and Nepal and recognised in the Eighth Schedule of the Indian Constitution, yet it has very few annotated NLP resources. The authors describe a manually POS-annotated Maithili corpus of roughly 25,000 sentences from several domains, including administration, law, education, health and tourism. Annotators worked in the web-based ILCIANN tool with the hierarchical BIS-based tagset used for Maithili, which has 11 top-level categories such as N, PR, DM, V, RP, QT and RD. The paper does not train or evaluate a tagger. Instead, it qualitatively analyses where annotation goes wrong:

  • English Penn Treebank habits carried over by annotators
  • intensifiers labelled as adverbs
  • confusion between light and main verbs
  • nouns inside conjunct verbs tagged as verbs
  • polysemous items like के, लेल and क
  • enclitic particles
  • honorific agreement
  • reduplication, which has no dedicated tag

The authors conclude that tagsets borrowed from Hindi or English miss these phenomena. They call for a Maithili-specific tagset, making the paper useful as a resource description and an error taxonomy for future Maithili tools.

Key points

  • Data per domain (Table 1): Administration 7000, Education 6310, Law 5267, Health 1008, Tourism 838, Agriculture 763, Technology 567.
    • These figures add up to about 21.7k sentences, not the stated 25k.
    • The abstract names only three of the seven domains.
  • Annotation setup: The ILCIANN tool (Kumar et al., 2012) combines manual tagging with partial automatic tagging of closed classes. It uses hierarchical BIS labels such as N_NN, V_VAUX, RP_INTF and RD_ECH (echo words).
  • No quantitative evaluation: The paper reports no tagging accuracy, no inter-annotator agreement and no UD mapping or LAS. Its findings come from illustrative examples only.
  • Typical errors cited:
    • बड tagged RB instead of RP_INTF.
    • भारतीय inside a party name tagged JJ, although it functions as part of a proper noun.
    • In मारि देब, the light verb देब risks being read as a second main verb.
  • Proposed extensions:
    • An HON feature, e.g. PRON[+HON] and VERB[+HON].
    • VOC.PART or ADDR tags for address words such as यौ, हौ, रौ, रे, which are currently lumped with interjections.
    • A reduplication tag.
    • A fully Maithili-specific tagset.

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

🎓Jawaharlal Nehru UniversityNew Delhi · India3 works · 170 citations

 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 2Research subjects and languages the work is about.Part-of-speech tagging · 126 worksLow-resource languages · 267 works

Cite

Cite as Shivani Priya, Shruti Jha, Urmila Jha, Girish Nath Jha, Deepali Tiwari and Jyoti Raj. 2026. POS Tagging in Low-Resource Maithili Language: Specific Challenges and Nuances. Proceedings of the 8th Workshop on Indian Language Data: Resources and Evaluation.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/27ugx7nj4vvs

@inproceedings{priya2026pos,
  title = {{POS Tagging in Low-Resource Maithili Language: Specific Challenges and Nuances}},
  author = {Shivani Priya, Shruti Jha, Urmila Jha, Girish Nath Jha, Deepali Tiwari and Jyoti Raj},
  booktitle = {Proceedings of the 8th Workshop on Indian Language Data: Resources and Evaluation},
  year = {2026},
  doi = {10.63317/27ugx7nj4vvs},
  url = {https://doi.org/10.63317/27ugx7nj4vvs}
}

References

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

  1. Bureau of Indian Standards (BIS). (2011). POS Tagset Guidelines for Indian Languages. find
  2. LDC-IL Maithili Speech Corpus Documentation. LDC-IL. Choudhary, N., & Ramamoorthy, L. (2019). LDC-IL Raw Text Corpora: An Overview. In Language Data Consortium for Indian Languages (LDC-IL). find
  3. Gopal, Madhav, & Jha, Girish Nath (2011). Tagging Sanskrit Using BIS POS Tagset. In Information Systems for Indian Languages: International Conference, ICISIL, Communications and Computer and Information Science, Springer, Volume-139, PP. 191-194. find
  4. Hardie, A., Archer, D., McEnery, T., Rayson, P. (2003). Developing an Automated Semantic Analysis System for Early Modern English. In Proceedings of Corpus Linguistics, PP. 22-31. find
  5. Jha, A. K., Singh, P. P., & Dwivedi, P. (2019, July). The Maithili text-to-speech system. In IEEE International Conference on Electronics, Computing and Communication Technologies (CONECCT 2019), PP. 1-6. IEEE. find
  6. Jha, Girish Nath and Kumari, Sangita (2010). Maithili Bhāṣā. In Bharatīya Bhāṣā Paricaya, Volume-2, PP. 453-478. Jha, S.K., Singh, P. P., & Kaul, V. K. find
  7. (2018). VEA Model. In Word Formation Process of Maithili MT. find
  8. Kumar, Ritesh, Kaushik, Shiv, Nainwani, Pinkey, Banerjee, Esha, Hadke, Sumedh, Jha, Girish Nath. (2012). Using the ILCI Annotation Tool for POS Annotation: A Case of Hindi. In IJCLA Vol.3, Number 2, Jul-Dec 2012, PP. 93-104. find
  9. Kumar, S. (2020). Named Entity Recognition in Maithili. find
  10. Banaras Hindu University, Varanasi. Kumar, Shantanu, Choudhary, Narayan (2025). Maithili Language Technology: A Survey. In Language in India, Volume-25(6), PP. 160-175. find
  11. Mundotiya, R. K., Gatla, Praveen, Kanwar, Nikita, Singh, Anil Kumar. (2025). Deep Learning-Based Similar Languages' POS Tagging: Experiments on Bhojpuri, Maithili, and Magahi. In Lecture Notes in Networks and Systems, Springer. find
  12. Priyadarshi, A., & Saha, S. K. (2023). A study on the performance of recurrent neural network-based models in Maithili part-of-speech tagging. In ACM Transactions on Asian and Low-Resource Language Information Processing. find
  13. Priyadarshi, A., & Saha, S. K. (2023). A study on the performance of recurrent neural network-based models in Maithili part-of-speech tagging. In ACM Transactions on Asian and Low-Resource Language Information Processing. find
  14. Singh, Srishti (2015), Challenges in Automatic POS Tagging of Indian Languages- A Comparative Study of Hindi and Bhojpuri, Master of Philosophy, Centre for Linguistics, Jawaharlal Nehru University. find
  15. Singh, Srishti, Banerjee, E. (2014). Annotating Bhojpuri Corpus using BIS Scheme. In LREC 2014 workshop WILDRE. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/priya-2026-pos-tagging-low-resource-maithili/