Elamur Research⌕
☰
home / papers / 2026 / TeamV at LT-EDI 2026: Multilingual Hate Speech Span Detection and Counter-Narrative Generation via Few-Shot In-Context Learning

TeamV at LT-EDI 2026: Multilingual Hate Speech Span Detection and Counter-Narrative Generation via Few-Shot In-Context Learning

Paper · 2026 · Vinay Babu Ulli and Jyoti Kumari

Summary

TeamV took part in the LT-EDI 2026 shared task on homophobic and transphobic social media comments. The task had two parts: finding and labelling hateful spans at the character level in English, Tamil and Hindi, and writing counter-narratives in English and Tamil. The authors did not train language-specific models. Instead, they prompted the proprietary Qwen3-Max model through the OpenRouter API with hand-picked demonstrations: 10 for span detection and 5 for generation, with Tamil examples swapped in for Tamil inputs.

LLMs are unreliable at producing character offsets, so the model also returned the span text itself. This text was then aligned to the original comment through a cascade of exact, normalized, case-insensitive and Levenshtein-based fuzzy matching.

The system ranked first in span detection for all three languages and won English counter-narrative generation. In Tamil generation it placed only fifth, producing polite but repetitive output. The work shows that prompting combined with careful post-processing can compete with dedicated architectures on multilingual span tasks. It also exposes weak generation quality in low-resource Dravidian languages.

Key points

  • Task 1 (macro F1, 1st place in every language): 0.5338 for English, 0.5272 for Tamil and 0.5478 for Hindi. Weighted F1 peaked at 0.6607 for Tamil.
  • Task 2, English (1st place): Overall Average was 87.47%, with CCNC 93.94%, QS 90.15% and PRS 90.91%.
  • Task 2, Tamil (5th place): The overall score was 64.30%. Distinct-2 was only 25.61%, compared with 73.56% for English, and QS (55.50%) and CCNC (66.51%) were low.
  • Span-matching ablation: The fallback matching stages recovered roughly 12–15% of valid spans that exact matching or raw indices would have lost.
  • Setup:
    • Decoding used temperature 0.3 and a limit of 512 output tokens.
    • Internal baselines were Gemma-3-12B-IT and Qwen3-8B-Instruct, tested zero-shot and with QLoRA.
    • Scripts and prompts are released at github.com/vinayulli/lt-edi-sharedtask.
    • The authors acknowledge that relying on a cloud API limits offline use and raises data-privacy concerns.

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

🎓Banaras Hindu UniversityVaranasi · India6 works · 260 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 3Research subjects and languages the work is about.Sentiment & emotion · 972 worksMultilingual & cross-lingual NLP · 828 worksBias, fairness & safety · 517 works

Cite

Cite as Vinay Babu Ulli and Jyoti Kumari. 2026. TeamV at LT-EDI 2026: Multilingual Hate Speech Span Detection and Counter-Narrative Generation via Few-Shot In-Context Learning. Proceedings of the Sixth Workshop on Language Technology for Equality, Diversity, Inclusion.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.ltedi-1.30

@inproceedings{ulli2026teamv,
  title = {{TeamV at LT-EDI 2026: Multilingual Hate Speech Span Detection and Counter-Narrative Generation via Few-Shot In-Context Learning}},
  author = {Vinay Babu Ulli and Jyoti Kumari},
  booktitle = {Proceedings of the Sixth Workshop on Language Technology for Equality, Diversity, Inclusion},
  year = {2026},
  doi = {10.18653/v1/2026.ltedi-1.30},
  url = {https://doi.org/10.18653/v1/2026.ltedi-1.30}
}

References

9 references, 4 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. 2024. Detection of homophobia and transphobia in youtube comments. International Journal of Data Science and Analytics, 18(1):49–68. find
  2. Bharathi Raja Chakravarthi, Ruba Priyadharshini, Thenmozhi Durairaj, John Philip McCrae, Paul Buitelaar, Prasanna Kumaresan, and Rahul Ponnusamy. 2022. Overview of the shared task on homophobia and transphobia detection in social media comments. In Proceedings of the Second Workshop on Language Tech… find
  3. Yi-Ling Chung, Elizaveta Kuzmenko, Serra Sinem Tekiroğlu, and Marco Guerini. 2019. Conan-counter narratives through nichesourcing: a multilingual dataset of responses to fight online hate speech. In Proceedings of the 57th annual meeting of the association for computational linguistics, pages 2819–…
  4. Assistant (Example 1): {"class": "Homophobia", "span_text": "[Extracted Span]", "positions":[start, end]} ... (followed by 9 more diverse examples) A.2 Margherita Fanton, Helena Bonaldi, Serra Sinem Tekiroğlu, and Marco Guerini. 2021. Human-in-the-loop for data collection: a multi-target counter na…
  5. Prasanna Kumar Kumaresan, Devendra Deepak Kayande, Ruba Priyadharshini, Paul Buitelaar, and Bharathi Raja Chakravarthi. 2025. Homophobia and transphobia span identification in low-resource languages. Natural Language Processing Journal, page 100169. find
  6. Prasanna Kumar Kumaresan, Praveen Prasannan, Tanay Singh, Ruba Priyadharshini, Subalalitha Chinnaudayar Navaneethakrishnan, Saranya Rajiakodi, Paul Buitelaar, and Bharathi Raja Chakravarthi. 2026. Findings of the Shared Task on Counter-Narrative Generation on Homophobic and Transphobic Comments. In… find
  7. Praveen Prasannan, Prasanna Kumar Kumaresan, Saranya Rajiakodi, CN Subalalitha, and Bharathi Raja Chakravarthi. 2025. Counter-speech generation for homophobic and transphobic social media content in malayalam. Social Network Analysis and Mining, 15(1):87. find
  8. Serra Sinem Tekiroğlu, Yi-Ling Chung, and Marco Guerini. 2020. Generating counter narratives against online hate speech: Data and strategies. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 1177–1190.
  9. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, and 1 others. 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/ulli-2026-teamv-lt-edi-2026-multilingual/