Elamur Research⌕
☰
home / papers / 2026 / Hi-SEMFLOW: Lie Algebra–Based Semantic Flow for Span-Level Informal Language Identification in Hindi

Hi-SEMFLOW: Lie Algebra–Based Semantic Flow for Span-Level Informal Language Identification in Hindi

Paper · 2026 · Manikandan Ravikiran, Tanmay Tiwari, Vibhu Gupta et al.

Summary

The paper addresses span-level detection of informal Hindi, such as slang and idioms, framed as BIO sequence labeling. Standard transformer token classifiers decide each token largely in isolation. Their output spans are often broken or contain illegal tag sequences, such as an I-INF tag directly after O. CRFs and constrained decoding can fix this, but they add discrete inference and rigid rules. The authors instead propose Hi-SEMFLOW, a differentiable layer placed on top of an encoder's logits. At each position, a projection head mixes a small set of learned antisymmetric generator matrices into a context-dependent operator. The layer exponentiates this operator, applies it to the previous token's logits, and blends the result with the current token's own logits. Training uses cross-entropy plus a flow regularizer and a curriculum on refinement strength. On HiSlang-4.9k, the method cuts invalid BIO transitions by as much as half and raises span-level F1 by 2–3 points. This shows that soft geometric refinement can replace hard structured decoding.

Key points

  • Benchmark and encoders: Evaluation uses the HiSlang-4.9k benchmark (Tiwari et al., 2025) with three labels (B-INF, I-INF, O). Both Hindi-specific and multilingual pretrained encoders are tested.
  • Method details:
    • The exponential of the combined generators is approximated by a third-order Taylor expansion.
    • Propagation runs left-to-right from the base logits rather than the refined ones, so structural bias does not compound over long spans.
    • A strength parameter λ weighs lexical evidence against structural evidence.
  • Main results:
    • Invalid BIO transitions fall by up to 50%.
    • Span-level F1 rises by 2–3 absolute points.
    • Macro-F1 gains are consistent but modest for Hindi-pretrained models, and depend on the encoder.
  • Ablations and errors:
    • Ablations show sensitivity to the number of generators, curriculum strength and interaction with decoding constraints.
    • Overly strong refinement can slightly over-extend spans, though the resulting tag sequences remain valid BIO.
  • Limitations and availability:
    • Matrix exponentiation adds computational cost, which may become a problem for larger label sets.
    • Evaluation covers only binary informal-span identification in Hindi.
    • Data and code are available upon request from the authors.

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

🎓Indian Institute of Technology MandiMandi · India2 works · 3 citations

 Fields, topics and concepts

📚 Shelves 5The 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 worksComputer Vision and Pattern Recognition · 2,869 works

Cite

Cite as Manikandan Ravikiran, Tanmay Tiwari, Vibhu Gupta and Rohit Saluja. 2026. Hi-SEMFLOW: Lie Algebra–Based Semantic Flow for Span-Level Informal Language Identification in Hindi. 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 10.63317/2duxp4m72ddj

@inproceedings{ravikiran2026hi,
  title = {{Hi-SEMFLOW: Lie Algebra–Based Semantic Flow for Span-Level Informal Language Identification in Hindi}},
  author = {Manikandan Ravikiran, Tanmay Tiwari, Vibhu Gupta and Rohit Saluja},
  booktitle = {Proceedings of the Second workshop on Challenges in Processing South Asian Languages (CHiPSAL2026)},
  year = {2026},
  doi = {10.63317/2duxp4m72ddj},
  url = {https://doi.org/10.63317/2duxp4m72ddj}
}

References

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

  1. SM Archana, Jay Prakash, Pramod Kumar Singh, and Waquar Ahmed. 2023. An effective biomedical named entity recognition by handling imbalanced data sets using deep learning and rule-based methods. SN Computer Science, 4(5):650. find
  2. David Belanger and Andrew McCallum. 2016. Structured prediction energy networks. In Proceedings of The 33rd International Conference on Machine Learning (ICML), volume 48 of Proceedings of Machine Learning Research, pages 983–992, New York, New York, USA. PMLR. find
  3. Michael M. Bronstein, Joan Bruna, Taco Cohen, and Petar Velivckovi’c. 2021. Geometric deep learning: Grids, groups, graphs, geodesics, and gauges. ArXiv, abs/2104.13478. — [arXiv:2104.13478](https://arxiv.org/abs/2104.13478) find
  4. Taco Cohen and Max Welling. 2016. Group equivariant convolutional networks. In Proceedings of The 33rd International Conference on Machine Learning (ICML), volume 48 of Proceedings of Machine Learning Research, pages 2990–2999, New York, New York, USA. PMLR. find
  5. Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2020. Unsupervised cross-lingual representation learning at scale. In Proceedings of ACL. find
  6. Leyang Cui and Yue Zhang. 2019. Hierarchically-refined label attention network for sequence labeling. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), pages 4115–4128. find
  7. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of NAACL-HLT, pages 4171–4186. Association for Computational Linguistics.
  8. Jacob Eisenstein. 2013. What to do about bad language on the internet. In Proceedings of the 2013 conference of the North American Chapter of the association for computational linguistics: Human language technologies, pages 359–369.
  9. Luca Falorsi, Pim de Haan, Tim R. Davidson, and Patrick Forré. 2019. Reparameterizing distributions on Lie groups. In Proceedings of the 22nd International Conference on Artificial Intelligence and Statistics (AISTATS), volume 89 of Proceedings of Machine Learning Research, pages 3244–3253. PMLR. find
  10. Jan E. Gerken, Jimmy Aronsson, Oscar Carlsson, Hampus Linander, Fredrik Ohlsson, Christoffer Petersson, and Daniel Persson. 2023. Geometric deep learning and equivariant neural networks. Artificial Intelligence Review, 56(12):14605–14662. find
  11. Divyanshu Kakwani, Anoop Kunchukuttan, Satish Golla, Nikhil Gokul, Pushpak Bhattacharyya, Mitesh M. Khapra, and Pratyush Kumar. 2020. Indicnlpsuite: Monolingual corpora, evaluation benchmarks and pre-trained multilingual language models for indian languages. In Findings of EMNLP.
  12. Imed Keraghel, Stanislas Morbieu, and Mohamed Nadif. 2024. Recent advances in named entity recognition: A comprehensive survey and comparative study. find
  13. Simran Khanuja, Diksha Bansal, Shashank Mehtani, Sarthak Khosla, Anshuman Dey, Naman Goyal, Parag Jain, Preksha Meena, Shubham Sharma, Karthik Sankaranarayanan, et al. 2021. Muril: Multilingual representations for indian languages. In Findings of EMNLP.
  14. John Lafferty, Andrew McCallum, and Fernando CN Pereira. 2001. Conditional random fields: Probabilistic models for segmenting and labeling sequence data. find
  15. Brian Lester, Daniel Pressel, Amy Hemmeter, Sagnik Ray Choudhury, and Srinivas Bangalore. 2020. Constrained decoding for computationally efficient named entity recognition taggers. arXiv preprint arXiv:2010.04362. find
  16. Xuezhe Ma and Eduard Hovy. 2016. End-to-end sequence labeling via bi-directional lstm-cnns-crf. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1064–1074.
  17. Rudra Murthy, Pallab Bhattacharjee, Rahul Sharnagat, Jyotsana Khatri, Diptesh Kanojia, and Pushpak Bhattacharyya. 2022. Hiner: A large hindi named entity recognition dataset. arXiv preprint arXiv:2204.13743. find
  18. Badry Ali Mustofa and Wawan Laksito Yuly Saptomo. 2025. Use of natural language processing in social media text analysis. Journal of Artificial Intelligence and Engineering Applications (JAIEA), 4(2):1235–1238. find
  19. Zhengqi Pei, Zhewei Sun, and Yang Xu. 2019. Slang detection and identification. In Proceedings of the 23rd conference on computational natural language learning (CoNLL), pages 881– 889. find
  20. Q-Success. 2024. Usage of hindi broken down by content management systems. Accessed: 2025-06-08. find
  21. Lance Ramshaw and Mitch Marcus. 1995. Text chunking using transformation-based learning. In Third workshop on very large corpora.
  22. Lev Ratinov and Dan Roth. 2009. Design challenges and misconceptions in named entity recognition. In Proceedings of the thirteenth conference on computational natural language learning (CoNLL-2009), pages 147–155.
  23. Manikandan Ravikiran, Tanmay Tiwari, Vibhu Gupta, Rakesh Prakash, Rohit Saluja, and Shayan Mohanty. 2026. Do tokenizers fail on informal hindi expressions? evidence from static, downstream, and robustness analyses. In Proceedings of the Second Workshop on Language Models for Low-Resource Languages (…
  24. Vinay Singh, Deepanshu Vijay, Syed Sarfaraz Akhtar, and Manish Shrivastava. 2018. Named entity recognition for hindi-english code-mixed social media text. In Proceedings of the seventh named entities workshop, pages 27–35. find
  25. Zhewei Sun, Qian Hu, Rahul Gupta, Richard Zemel, and Yang Xu. 2024. Toward informal language processing: Knowledge of slang in large language models. arXiv:2404.02323. arXiv preprint Simone Tedeschi, Federico Martelli, and Roberto Navigli. 2022. Id10m: Idiom identification in 10 languages. In Findin…
  26. Tanmay Tiwari, Vibhu Gupta, Manikandan Ravikiran, and Rohit Saluja. 2025. Hislang-4.9 k: A benchmark dataset for hindi slang detection and identification. In Proceedings of the 8th International Conference on Natural Language and Speech Processing (ICNLSP-2025), pages 464–472. find
  27. Yogarshi Vyas, Spandana Gella, Jatin Sharma, Kalika Bali, and Monojit Choudhury. 2014. Pos tagging of english-hindi code-mixed social media content. In Proceedings of the 2014 conference on empirical methods in natural language processing (EMNLP), pages 974–979.
  28. Shuai Zheng, Sadeep Jayasumana, Bernardino Romera-Paredes, Vibhav Vineet, Zhizhong Su, Dalong Du, Chang Huang, and Philip H. S. Torr. 2015. Conditional random fields as recurrent neural networks. 2015 IEEE International Conference on Computer Vision (ICCV), pages 1529–1537. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/ravikiran-2026-hi-semflow-lie-algebra-based/