Elamur Research⌕
☰
home / papers / 2026 / From Romanized to Devanagari: Enhancing Nepali Sentiment Analysis with NepaliXlit

From Romanized to Devanagari: Enhancing Nepali Sentiment Analysis with NepaliXlit

Paper · 2026 · Suraj Patel, Kashish Kumari Dhami, Norden Sherpa et al.

Summary

On social media, Nepali is mostly typed in Latin letters with no fixed spelling. The same word can appear as "chha", "cha", or "xa". Most Nepali-capable and multilingual models, however, were pre-trained on Devanagari text. To close this script gap, the authors adapt IndicXlit, which was built largely from formal dictionary word pairs, into NepaliXlit. This transliterator is fine-tuned on informal Romanized–Devanagari word pairs taken from social media. To test whether better transliteration helps a downstream task, they collected over 6,500 Romanized Nepali comments and manually labelled a balanced three-class sentiment subset. Fine-tuned mBERT and MuRIL do poorly on raw Romanized input but improve once the comments are converted to Devanagari with NepaliXlit. Zero-shot LLMs read the Romanized text directly and beat every encoder setup. The study contributes an open tool and dataset for an under-resourced language. It shows that transliteration is a worthwhile bridge for small, cheap encoder models, while frontier LLMs can largely skip this preprocessing step.

Key points

  • NepaliXlit was fine-tuned from IndicXlit on 2,943 informal word pairs and tested on 736 held-out pairs. It raises transliteration accuracy by 8% and lowers character error rate by 11%.
  • The new sentiment resource has more than 6,500 Romanized comments, including a balanced, manually annotated subset of 1,518 instances (positive / neutral / negative). Code-mixed Nepali-English text was excluded.
  • Converting input to Devanagari with NepaliXlit consistently improved classification accuracy for both mBERT and MuRIL.
  • In zero-shot mode, Gemini-2.5-Pro surpassed MuRIL, the strongest of the fine-tuned encoders, with a margin above 13 percentage points. GPT variants and Qwen3-Max were also evaluated.
  • The neutral class was the hardest for every model and script. The authors list as limitations the small dataset, zero-shot-only LLM prompting, and the lack of a measurement of token fragmentation. Data and code are released on GitHub.

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

🎓Birmingham City UniversityBirmingham · United Kingdom5 works · 26 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 1Research subjects and languages the work is about.Sentiment & emotion · 972 works

Cite

Cite as Suraj Patel, Kashish Kumari Dhami, Norden Sherpa and Supriya Khadka. 2026. From Romanized to Devanagari: Enhancing Nepali Sentiment Analysis with NepaliXlit. Proceedings of the Second workshop on Challenges in Processing South Asian Languages (CHiPSAL2026).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/5j8kkeb2myf6

@inproceedings{patel2026romanized,
  title = {{From Romanized to Devanagari: Enhancing Nepali Sentiment Analysis with NepaliXlit}},
  author = {Suraj Patel, Kashish Kumari Dhami, Norden Sherpa and Supriya Khadka},
  booktitle = {Proceedings of the Second workshop on Challenges in Processing South Asian Languages (CHiPSAL2026)},
  year = {2026},
  doi = {10.63317/5j8kkeb2myf6},
  url = {https://doi.org/10.63317/5j8kkeb2myf6}
}

References

40 references, 10 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Sandhini Agarwal, Lama Ahmad, Jason Ai, Sam Altman, Andy Applebaum, Edwin Arbus, Rahul K Arora, Yu Bai, Bowen Baker, Haiming Bao, et al. 2025. gpt-oss-120b & gpt-oss-20b model card. arXiv preprint arXiv:2508.10925.
  2. Tess Anderson, Sayani Sarkar, and Robert Kelley. 2024. Analyzing public sentiment on sustainability: A comprehensive review and application of sentiment analysis techniques. Natural Language Processing Journal, 8:100097. find
  3. Gulfarogh Azam, Mohd Sadique, Saif Ali, Mohammad Nadeem, Erik Cambria, Shahab Saquib Sohail, and Mohammad Sultan Alam. 2025. Beyond specialization: Benchmarking llms for transliteration of indian languages. arXiv preprint arXiv:2505.19851. — [arXiv:2505.19851](https://arxiv.org/abs/2505.19851) find
  4. Dipesh Baral. 2025. Comparing labse with contrastively and soft-label fine-tuned mbert models for semantic search over a nepali knowledge base. International Journal on Engineering Technology, 3(1):146–155. find
  5. Francesco Barbieri, Luis Espinosa Anke, and Jose Camacho-Collados. 2022. Xlm-t: Multilingual language models in twitter for sentiment analysis and beyond. In Proceedings of the thirteenth language resources and evaluation conference, pages 258–266.
  6. Hemanta Baruah, Sanasam Ranbir Singh, and Priyankoo Sarmah. 2024. Assamesebacktranslit: Back transliteration of romanized assamese social media text. In Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024), pages 16… find
  7. Deepti Bhalla, Nisheeth Joshi, and Iti Mathur. 2013. Rule based transliteration scheme for english to punjabi. arXiv preprint arXiv:1307.4300. — [arXiv:1307.4300](https://arxiv.org/abs/1307.4300) find
  8. Chandra Prakash Chaudahry, Basanta Joshi, Aman Shakya, and Santosh Giri. 2025. Neptun: Normalization for romanized nepali sentiment analysis. Journal of Himalaya College of Engineering, 2(1). find
  9. Gheorghe Comanici, Eric Bieber, Mike Schaekermann, Ice Pasupat, Noveen Sachdeva, Inderjit Dhillon, Marcel Blistein, Ori Ram, Dan Zhang, Evan Rosen, et al. 2025. Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities. arXiv prep…
  10. Ram Kumar Dahal. 2000. Language polics innepal. Contributions to Nepalese Studies, 27(2):155– 190. find
  11. Yomal De Mel, Kasun Wickramasinghe, Nisansa De Silva, and Surangika Ranathunga. 2025. Sinhala transliteration: a comparative analysis between rule-based and seq2seq approaches. In Proceedings of the First Workshop on Natural Language Processing for Indo-Aryan and Dravidian Languages, pages 166–173. find
  12. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies,…
  13. George Forman. 2008. Bns feature scaling: an improved representation over tf-idf for svm text classification. In Proceedings of the 17th ACM conference on Information and knowledge management, pages 263–270. find
  14. Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, et al. 2024. The llama 3 herd of models. arXiv preprint arXiv:2407.21783.
  15. Ehtesham Hashmi, Sule Yildirim Yayilgan, Ibrahim A Hameed, Muhammad Mudassar Yamin, Mohib Ullah, and Mohamed Abomhara. 2024. Enhancing multilingual hate speech detection: From language-specific insights to cross-linguistic integration. IEEE Access, 12:121507–121537. find
  16. J Jaavid, Raj Dabre, M Aswanth, Jay Gala, Thanmay Jayakumar, Ratish Puduppully, and Anoop Kunchukuttan. 2024. Romansetu: Efficiently unlocking multilingual capabilities of large language models via romanization. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguisti… find
  17. Vani Kanjirangat, Tanja Samardzic, Ljiljana Dolamic, and Fabio Rinaldi. 2025. Tokenization and representation biases in multilingual models on dialectal nlp tasks. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 24003–24021. find
  18. Pilot Khadka, Ankit Bk, Ashish Acharya, Bikram Kc, Sandesh Shrestha, and Rabin Thapa. 2025. Nepali transformers@ nlu of devanagari script languages 2025: Detection of language, hate speech and targets. In Proceedings of the First Workshop on Challenges in Processing South Asian Languages (CHiPSAL 20… find
  19. Supriya Khadka and Bijayan Bhattarai. 2025. Gender bias in nepali-english machine translation: A comparison of llms and existing mt systems. In Proceedings of the 6th Workshop on Gender Bias in Natural Language Processing (GeBNLP), pages 75–82. find
  20. Supriya Khadka, Ranju G.C., Prabin Paudel, Rahul Shah, and Basanta Joshi. 2023. Nepali text-to-speech synthesis using tacotron2 for mel-spectrogram generation. In SIGUL 2023, 2nd Annual Meeting of the Special Interest Group on Under-resourced Languages: a Satellite Workshop of Interspeech 2023. find
  21. Samir Khanal, Ranjan Paudel, and Rajendra Chataut. 2021. Nepali printed text to speech with accurate transliterated form. International Journal of Computer Science and Information Technologies (IJCSIT), 12(5):97–101. find
  22. Simran Khanuja, Diksha Bansal, Sarvesh Mehtani, Savya Khosla, Atreyee Dey, Balaji Gopalan, Dilip Kumar Margam, Pooja Aggarwal, Rajiv Teja Nagipogu, Shachi Dave, et al. 2021. Muril: Multilingual representations for indian languages. arXiv preprint arXiv:2103.10730.
  23. Uttam Kumar, Tanusree Das, Raju S Bapi, Prakash Padakannaya, R Malatesha Joshi, and Nandini C Singh. 2010. Reading different orthographies: an fmri study of phrase reading in hindi–english bilinguals. Reading and Writing, 23(2):239–255. find
  24. Yash Madhani, Mitesh M Khapra, and Anoop Kunchukuttan. 2023a. Bhasa-abhijnaanam: Native-script and romanized language identification for 22 indic languages. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 816–826. find
  25. Yash Madhani, Sushane Parthan, Priyanka Bedekar, Gokul Nc, Ruchi Khapra, Anoop Kunchukuttan, Pratyush Kumar, and Mitesh M Khapra. 2023b. Aksharantar: Open indic-language transliteration datasets and models for the next billion users. In Findings of the association for computational linguistics: Emnl… find
  26. Gagan Reddy Narayanaswamy. 2021. Exploiting BERT and RoBERTa to Improve Performance for Aspect Based Sentiment Analysis. Dissertation, Technological University Dublin, Dublin, Ireland. find
  27. Myle Ott, Sergey Edunov, Alexei Baevski, Angela Fan, Sam Gross, Nathan Ng, David Grangier, and Michael Auli. 2019. fairseq: A fast, extensible toolkit for sequence modeling. In Proceedings of the 2019 conference of the North American chapter of the association for computational linguistics (Demonstr…
  28. Abhash Pradhananga and Anand Kumar Sah. 2023. Transformer-based deep learning models for sentiment analysis in romanized nepali: a comparative investigation of bert and roberta. In Proceedings of 14th IOE graduate conference, volume 14. language model for the nepali language. arXiv preprint arXiv:25… — [arXiv:2506.16399](https://arxiv.org/abs/2506.16399) find
  29. Irina Rish et al. 2001. An empirical study of the naive bayes classifier. In IJCAI 2001 workshop on empirical methods in artificial intelligence, volume 3, pages 41–46. Seattle, USA. find
  30. Phillip Rust, Jonas Pfeiffer, Ivan Vulić, Sebastian Ruder, and Iryna Gurevych. 2021. How good is your tokenizer? on the monolingual performance of multilingual language models. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint…
  31. Koyyalagunta Krishna Sampath and M Supriya. 2024. Transformer based sentiment analysis on code mixed data. Procedia Computer Science, 233:682–691. find
  32. Pintu Sarkar. 2026. The evolution of the nepali language: A brief history. International Journal of Creative Research Thoughts (IJCRT), 14(1):a874–a880. Available at: http://www.ijcrt.org/papers/IJCRT2601109.pdf. ↗
  33. Roy Schwartz, Jesse Dodge, Noah A Smith, and Oren Etzioni. 2020. Green ai. Communications of the ACM, 63(12):54–63.
  34. Weijia Shi, Anirudh Ajith, Mengzhou Xia, Yangsibo Huang, Daogao Liu, Terra Blevins, Danqi Chen, and Luke Zettlemoyer. 2023. Detecting pretraining data from large language models. arXiv preprint arXiv:2310.16789. find
  35. Suresh Man Shrestha. 2012. Transliteration system for nepali language. Journal on Geoinformatics, Nepal, 11:37–41. find
  36. Sameer Sitoula, Tej Bahadur Shahi, Laxmi Prasad Bhatt, Anisha Pokhrel, and Arjun Neupane. 2025. Nepemo: A multi-label emotion and sentiment analysis on nepali reddit with linguistic insights and temporal trends. arXiv preprint arXiv:2512.22823. — [arXiv:2512.22823](https://arxiv.org/abs/2512.22823) find
  37. Shushanta Pudasaini, Sunil Ghimire, Prabhat Ale, Aman Shakya, Prakriti Paudel, and Basanta Joshi. 2024. Application of nepali large language models to improve sentiment analysis. In Proceedings of the 2024 7th International Conference on Computers in Management and Business, pages 144–150. find
  38. Sulav Timilsina, Milan Gautam, and Binod Bhattarai. 2022. Nepberta: Nepali language model trained in a large corpus. In Proceedings of the 2nd Conference of the Asia-Pacific Chapter of the Association for Computational Linguistics and the 12th International Joint Conference on Natural Language Proce… find
  39. Shushanta Pudasaini, Aman Shakya, Siddhartha Shrestha, Sahil Bhatta, Sunil Thapa, and Sushmita Palikhe. 2025. Nepaligpt: A generative Haoran Xu, Benjamin Van Durme, and Kenton Murray. 2021. Bert, mbert, or bibert? a study on contextualized embeddings for neural machine translation. In Proceedings of… find
  40. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, et al. 2025. Qwen3 technical report. arXiv preprint arXiv:2505.09388.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/patel-2026-romanized-devanagari-enhancing-nepali-sentiment/