Elamur Research⌕
☰
home / papers / 2026 / DuoNova@LTEDI 2026: Multilingual Span Detection and Counter-Narrative Generation on Homophobic and Transphobic Comments

DuoNova@LTEDI 2026: Multilingual Span Detection and Counter-Narrative Generation on Homophobic and Transphobic Comments

Paper · 2026 · Manasa S, Arohi Rawat and Anbukkarasi Sampath

Summary

This paper describes the DuoNova team's entry in the LT-EDI @ ACL 2026 shared task on online abuse aimed at LGBTQ+ people. The task had two parts. Task 1 asked systems to locate the homophobic or transphobic fragment inside a comment in English, Tamil and Hindi. Task 2 asked them to write a polite counter-narrative reply in English and Tamil. The organisers released 1,800 training and 66 test items. Each training item links a comment to its harmful span, a human-written response and an abuse label. Neutral tokens heavily outnumber harmful ones. The authors framed span detection as BIO token labelling and reply writing as prompted sequence-to-sequence generation with greedy decoding. They fine-tuned google/flan-t5-base for English and google/mt5-small for Tamil and Hindi. The system ranked second in Task 1 for all three languages, although macro F1 stayed around 0.5. In Task 2, Tamil replies scored as very polite but showed little lexical variety. The work shows that compact pretrained models give a usable baseline, but there are no ablations or baseline comparisons.

Key points

  • Models and settings: google/flan-t5-base for English and google/mt5-small for Tamil and Hindi. Both used a maximum length of 128, batch size 8, 3 epochs and a learning rate of 5e-5 with HuggingFace Transformers. Class imbalance was not reweighted.
  • Task 1 macro F1 scores were 0.5111 for English, 0.5090 for Tamil and 0.4585 for Hindi. Weighted F1 was 0.6490 for English and 0.6737 for Tamil. The system ranked second in every language.
  • Task 2 English results: BERTScore-F1 of 86.04%, Distinct-2 of 58.22% and an overall average of 57.79%. The rubric-based QS was only 37.88%.
  • Task 2 Tamil results: PRS (politeness/respect) of 94.50% and an overall average of 62.15%. Distinct-2 was just 3.62%, which points to repetitive outputs.
  • Error analysis found three main problems:
    • Grammatical slips in Tamil mT5 outputs.
    • Short or generic replies.
    • Missed spans when the hostility was implicit.
  • Code is available at github.com/Manasa-S-02/DuoNova.

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

🎓Manipal Academy of Higher EducationManipal · India5 works · 57 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 works

🧭 Topics 1Research subjects and languages the work is about.Multilingual & cross-lingual NLP · 882 works

Cite

Cite as Manasa S, Arohi Rawat and Anbukkarasi Sampath. 2026. DuoNova@LTEDI 2026: Multilingual Span Detection and Counter-Narrative Generation on Homophobic and Transphobic Comments. 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-1.17

@inproceedings{s2026duonova,
  title = {{DuoNova@LTEDI 2026: Multilingual Span Detection and Counter-Narrative Generation on Homophobic and Transphobic Comments}},
  author = {Manasa S, Arohi Rawat and Anbukkarasi Sampath},
  booktitle = {Proceedings of the Sixth Workshop on Language Technology for Equality, Diversity, Inclusion},
  year = {2026},
  doi = {10.18653/v1/2026.ltedi-1.17},
  url = {https://doi.org/10.18653/v1/2026.ltedi-1.17}
}

References

12 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Binny Mathew, Ritam Dutt, Pawan Goyal, and Animesh Mukherjee. 2019. Spread of hate speech in online social media. In Proceedings of the Web Conference. find
  2. Marzieh Mozafari, Reza Farahbakhsh, and Noel Crespi. 2019. A bert-based transfer learning approach for hate speech detection. In International Conference on Complex Networks. find
  3. 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
  4. Ruba Priyadharshini and Bharathi Raja Chakravarthi. 2021. Dravidian offensive language dataset in tamil, malayalam, and kannada. In Proceedings of the First Workshop on Speech and Language Technologies for Dravidian Languages. find
  5. Bharathi Raja Chakravarthi. 2024. Detection of homophobia and transphobia in youtube comments. International Journal of Data Science and Analytics, 18(1):49–68. find
  6. Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J. Liu. 2020. Exploring the limits of transfer learning with a unified text-to-text transformer. Journal of Machine Learning Research.
  7. Bharathi Raja Chakravarthi, Ruba Priyadharshini, Vigneshwaran Muralidaran, Elizabeth Sherly, and John McCrae. 2021. Overview of the dravidian-codemix 2021 shared task on offensive language identification. In Proceedings of the First Workshop on Speech and Language Technologies for Dravidian Language… find
  8. Linting Xue, Noah Constant, Adam Roberts, Mihir Kale, Rami Al-Rfou, Aditya Siddhant, Aditya Barua, and Colin Raffel. 2021. mt5: A massively multilingual pre-trained text-to-text transformer. In Proceedings of NAACL.
  9. Bharathi Raja Chakravarthi, Ruba Priyadharshini, and Elizabeth Sherly. 2020. Multilingual offensive language identification in dravidian languages. Information Processing and Management. find
  10. Yi-Ling Chung, Elizaveta Kuzmenko, Serra Sinem Tekiroğlu, and Marco Guerini. 2019. Conan – counter narratives through nichesourcing. In Proceedings of ACL. find
  11. 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.
  12. 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

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/s-2026-duonova-ltedi-2026-multilingual-span/