Elamur Research⌕
☰
home / papers / 2026 / Nepal Script Text Recognition from Ancient Artifacts: Challenges and Opportunities

Nepal Script Text Recognition from Ancient Artifacts: Challenges and Opportunities

Summary

Nepal Script, historically used for Nepal Bhasa, Sanskrit and other languages, is now considered endangered. Many inscriptions and manuscripts written in it remain undigitized, and there was no OCR ground truth for stone inscriptions. Building on their earlier CRNN-based recognizer for handwritten and printed text, the authors assembled a line-level dataset from several sources. They photographed stone inscriptions at 85 sites around the Kathmandu Valley. They added archival material from Tribhuvan University and the National Archives of Nepal, and included modern handwritten and typed documents. Experts verified the transcriptions, which use the Newa Unicode block. The recognition pipeline first finds text lines with YOLOv8, converts them to grayscale, and then transcribes them with a CRNN trained using CTC loss. After augmentation, the model reached a character error rate of 12.61%. It also produced usable output on unseen inscriptions and manuscripts, and even on wooden and copper plates. This makes the work a practical starting point for digitizing heritage material in Nepal Script and related Nepalese scripts.

Key points

  • The dataset contains 5,219 text-line images: 2,371 from stone inscriptions, 1,758 from religious manuscripts, and the rest from modern handwritten and typed documents. Augmentation expanded it to 41,752 samples.
  • The pipeline combines YOLOv8 for line detection with CRNN + CTC for recognition, achieving a CER of 12.61%. The authors' earlier system scored 6.65% CER and 13.11% WER, but only on handwritten and printed text.
  • The script is hard to recognize for several reasons: ancient texts rarely separate words, conjuncts are numerous, and many characters have variant forms. The header line also affects which vowel modifiers are used.
  • The released resources are the Nepal Script Stone Inscription Dataset on Kaggle (CC BY-NC-SA 4.0) and the source code on GitHub (GPLv3). The secondary archival data cannot be redistributed because of licensing.
  • Related resources cited include the manuscript ground truth of O'Neill and Hill (2022), built with Transkribus HTR, and a Ranjana script character dataset. HTR-VT is noted as a promising approach for settings with little data.

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 10.63317/427yvm8biop7

@inproceedings{nakarmi2026nepal,
  title = {{Nepal Script Text Recognition from Ancient Artifacts: Challenges and Opportunities}},
  author = {Nakarmi, Swornim and Sthapit, Sarin and Tuladhar, Sahil Ratna and Shakya, Arya and Bal, Bal Krishna and Chulyadyo, Rajani},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation Conference},
  year = {2026},
  doi = {10.63317/427yvm8biop7},
  url = {https://doi.org/10.63317/427yvm8biop7}
}

References

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

  1. Jen Bati and Pankaj Raj Dawadi. 2023. Ranjana script handwritten character recognition using cnn. JOIV: International Journal on Informatics Visualization, 7(3):984–990. find
  2. Alexey Dosovitskiy. 2020. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929. find
  3. Kartik Dutta, Praveen Krishnan, Minesh Mathew, and CV Jawahar. 2018. Offline handwriting recognition on devanagari using a new benchmark dataset. In 2018 13th IAPR international workshop on document analysis systems (DAS), pages 25–30. IEEE. find
  4. Agam Dwivedi, Rohit Saluja, and Ravi Kiran Sarvadevabhatla. 2020. An ocr for classical indic documents containing arbitrarily long words. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition Workshops, pages 560–561. find
  5. Himalayan News Service. March 14, 2021. Kmc makes nepal bhasa mandatory in schools. The Himalayan Times. Accessed on: October 14, 2025. find
  6. Yuting Li, Dexiong Chen, Tinglong Tang, and Xi Shen. 2025. Htr-vt: Handwritten text recognition with vision transformer. Pattern Recognition, 158:110967. find
  7. Ajoy Mondal and CV Jawahar. 2022. Enhancing indic handwritten text recognition using global semantic information. In International Conference on Frontiers in Handwriting Recognition, pages 360–374. Springer. find
  8. Swornim Nakarmi, Sarin Sthapit, Arya Shakya, Rajani Chulyadyo, and Bal Krishna Bal. 2024. Nepal script text recognition using crnn ctc architecture. In Proceedings of the 3rd Annual Meeting of the Special Interest Group on Under-resourced Languages@ LREC-COLING 2024, pages 244–251. find
  9. Nepal Lipi Guthi. 1992. Pracalit Nepāl Lipiyā Varṇamālā. Nepal Lipi Guthi. find
  10. Alexander James O’Neill and Nathan Hill. 2022. Text recognition for nepalese manuscripts in pracalit script. Journal of Open Humanities Data, 8. find
  11. Anshuman Pandey. 2012. Proposal to encode the newar script in iso/iec 10646. find
  12. Ram Chandra Pandey, Babu Ram Dawadi, Suman Sharma, and Abinash Basnet. 2017. Dictionary based nepali word recognition using neural network. Int. J. Sci. Eng. Res, pages 473–479. find
  13. Ashok Kumar Pant, Sanjeeb Prasad Panday, and Shashidhar Ram Joshi. 2012. Off-line nepali handwritten character recognition using multi-layer perceptron and radial basis function neural networks. In 2012 Third Asian Himalayas International Conference on Internet, pages 1–5. IEEE. find
  14. Shankarman Rajbanshi. 1970. Kantipur shilalekh suchi [kathmandu stone inscriptions list]. find
  15. Joseph Redmon, Santosh Divvala, Ross Girshick, and Ali Farhadi. 2016. You only look once: Unified, real-time object detection. In Proceedings of the IEEE conference on computer vision and pattern recognition, pages 779–788. find
  16. Bikash Shaw, Swapan Kumar Parui, and Malayap-pan Shridhar. 2008. Offline handwritten devanagari word recognition: A holistic approach based on directional chain code feature and hmm. In 2008 International Conference on Information Technology, pages 203–208. IEEE. find
  17. Bal Gopal Shrestha. 1999. The newars: The indigenous population of the kathmandu valley in the modern state of nepal. The Journal of Newar Studies, 2:1–16. find
  18. Brijmohan Singh, Ankush Mittal, MA Ansari, and Debashis Ghosh. 2011. Handwritten devanagari word recognition: a curvelet transform based approach. International Journal on Computer Science and Engineering, 3(4):1658– 1665. find
  19. Unicode, Inc. 2023. Newa Range: 11400-1147F. Accessed on: February 24, 2024. find
  20. Regent Nanda Vaidya and Bal Krishna Bal. 2024. Deep learning approach for offline nepali handwritten word recognition. In International Conference on Analysis of Images, Social Networks and Texts, pages 138–150. Springer. find
  21. Dhanavajra Vajracharya. 1999. Madhya kalka ab-hilekha [inscriptions of the medieval period]. find
  22. John Whelpton. 2005. A history of Nepal. Cambridge University Press. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/papers/nakarmi-2026-nepal-script-text-recognition-ancient/