Elamur Research⌕
☰
home / papers / 2026 / Does Hate Transfer? Cross-Lingual Generalisation of Offensive Content Detection Across Indic Languages

Does Hate Transfer? Cross-Lingual Generalisation of Offensive Content Detection Across Indic Languages

Paper · 2026 · Purandhar M. Reddy and Sara Renjit

Summary

The paper asks whether offensive-content detectors trained on one Indic language still work on a related language without new annotations. The authors fine-tuned LLaMA-3.1-8B-Instruct with LoRA separately on each of five languages in the MACD benchmark: Tamil, Telugu, Hindi, Kannada and Malayalam. MACD consists of near-balanced ShareChat comments. Each model was then run zero-shot on the other four test sets, which gives all twenty directed pairs. Performance drop was measured relative to the monolingual baseline. The authors also checked whether URIEL typological similarity (via lang2vec) accounts for the results.

Transfer mostly fails. Only three pairs stay under the 15% tolerable-loss threshold, and all three use Malayalam as the source. Results are strongly direction-dependent and do not follow typology. Different source languages also produce different kinds of errors. The practical message for low-resource moderation, including in other language families, is that membership in the same language family is no replacement for annotating each target language.

Key points

  • Baseline: in-language LoRA models average 0.858 macro-F1 (MCC 0.718). Telugu scores highest at 0.881.
  • Transfer outcomes:
    • 3 of 20 directed pairs fall below 15% loss.
    • Malayalam is the best source, with 16.8% average loss.
    • Telugu is the hardest target, with 33.8% average loss.
    • The worst pair is Tamil→Telugu, at F1 0.390 (49% loss).
  • Asymmetry: Malayalam→Tamil reaches F1 0.728, while Tamil→Malayalam reaches only 0.539.
  • No typological signal: URIEL similarities fall in a narrow band (0.840–0.939). Spearman ρ = −0.254 (p = 0.281), which is not significant.
  • Two failure modes:
    • Tamil- and Kannada-trained models under-flag, missing about 82% and 73% of offensive comments with only 2–5% false alarms.
    • Malayalam-trained models over-flag, missing about 39% but with roughly 21% false alarms.

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, topics and concepts

📚 Shelves 2The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 works

🧭 Topics 1Research subjects and languages the work is about.Multilingual & cross-lingual NLP · 828 works

Cite

Cite as Purandhar M. Reddy and Sara Renjit. 2026. Does Hate Transfer? Cross-Lingual Generalisation of Offensive Content Detection Across Indic Languages. Proceedings of the Second International Conference on Natural Language Processing and Artificial Intelligence for Cyber Security.

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{reddy2026does,
  title = {{Does Hate Transfer? Cross-Lingual Generalisation of Offensive Content Detection Across Indic Languages}},
  author = {Purandhar M. Reddy and Sara Renjit},
  booktitle = {Proceedings of the Second International Conference on Natural Language Processing and Artificial Intelligence for Cyber Security},
  year = {2026},
  url = {https://aclanthology.org/2026.nlpaics-1.24/}
}

References

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

  1. Bharathi Raja Chakravarthi, Ruba Priyadharshini, Navya Jose, Anand M. Kumar, Thomas Mandl, Prasanna Kumar Kumaresan, Rahul Ponnusamy, R. L. Hariharan, John P. McCrae, and Elizabeth Sherly. 2022. DravidianCodeMix: Sentiment analysis and offensive language identification dataset for Dravidian language… find
  2. Murray B. Emeneau. 1956. India as a linguistic area. Language, 32(1):3–16. find
  3. Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, et al. 2024. The Llama 3 herd of models. arXiv preprint arXiv:2407.21783.
  4. Vikram Gupta, Sumegh Roychowdhury, Mithun Das, Somnath Banerjee, Punyajoy Saha, Binny Mathew, Hastagiri P. Vanchinathan, and Animesh Mukherjee. find
  5. Edward J. Hu, Yelong Shen, Phillip Wallis, Zeyuan Allen-Zhu, Yuanzhi Li, Shean Wang, Lu Wang, and Weizhu Chen. 2022. LoRA: Low-rank adaptation of large language models. In Proceedings of the Tenth International Conference on Learning Representations.
  6. Junjie Hu, Sebastian Ruder, Aditya Siddhant, Graham Neubig, Orhan Firat, and Melvin Johnson. 2020. XTREME: A massively multilingual multi-task benchmark for evaluating cross-lingual generalisation. In Proceedings of the 37th International Conference on Machine Learning, pages 4411–4421. find
  7. Bhadriraju Krishnamurti. 2003. The Dravidian Languages. Cambridge Language Surveys. Cambridge University Press, Cambridge. find
  8. Patrick Littell, David R. Mortensen, Ke Lin, Katherine Kairis, Carlisle Turner, and Lori Levin. 2017. URIEL and lang2vec: Representing languages as typological, geographical, and phylogenetic vectors. In Proceedings of the 15th Conference of the European Chapter of the Association for Computational…
  9. Nedjma Ousidhoum, Zizheng Lin, Hongming Zhang, Yangqiu Song, and Dit-Yan Yeung. 2019. Multilingual and multi-aspect hate speech analysis. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processin…
  10. Telmo Pires, Eva Schlinger, and Dan Garrette. 2019. How multilingual is multilingual BERT? In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 4996–5001, Florence, Italy. Association for Computational Linguistics.
  11. Jimin Sun, Hwijeen Ahn, Chan Young Park, Yulia Tsvetkov, and David R. Mortensen. 2021. Cross-cultural similarity features for cross-lingual transfer learning of pragmatically motivated tasks. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguisti…
  12. Li Zhou, Antonia Karamolegkou, Wenyu Chen, and Daniel Hershcovich. 2023. Cultural compass: Predicting transfer learning success in offensive language detection with cultural features. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 12684–12702, Singapore. Association… find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/reddy-2026-does-hate-transfer-cross-lingual/