Elamur Research⌕
☰
home / papers / 2026 / Igniters@LTEDI 2026: Multilingual Gender-Inclusive Language Generation with mT5 and Counter-Narrative Generation Using Llama-3

Igniters@LTEDI 2026: Multilingual Gender-Inclusive Language Generation with mT5 and Counter-Narrative Generation Using Llama-3

Paper · 2026 · Rajendran S, Ramkumar N and Malarselvi R

Summary

The authors' submission to the LT-EDI ACL 2026 shared task tackles two problems linked to gender bias in text. The first is rewriting gender-marked sentences into neutral form in five languages: English, Tamil, Kannada, German and Spanish. The second is producing counter-narratives that push back on biased English statements. For rewriting, they fine-tuned mT5 with LoRA. They framed the task as minimal-edit sequence-to-sequence transformation and added a Chain-of-Thought-style instruction that asks the model to work out the context before swapping in an inclusive term. For counter-narratives, they gave Llama-3 8B a structured prompt that asks for a short, empathetic and non-insulting reply. An LLM-as-a-Judge framework scored both systems on rubric dimensions. Counter-narrative generation scored very high. Rewriting quality depended heavily on the language: Tamil and Spanish did best, while German and English trailed, mostly because their quality and relevance scores were weak. The study shows that a fairly lightweight setup on one GPU can serve inclusive-language tasks, including in morphologically rich Dravidian languages. However, the paper reports no baselines, ablations or human evaluation.

Key points

  • Data: The LT-EDI 2026 shared task data has 7,458 pairs in total. Subtask A contains 2,308 gender-neutral word pairs and 4,424 sentence pairs across five languages; German has only sentence pairs. Subtask B contains 726 English counterfactual pairs.
  • Setup: Training used Hugging Face Transformers on a single RTX 3090. mT5 was trained with batch size 4, AdamW, a learning rate of 5×10⁻⁵ and 5 epochs. Llama-3 8B decoding used temperature 0.7, top-p 0.9, top-k 50 and a repetition penalty of 1.1.
  • Subtask A averages (mean of GA, GN and QR): Tamil 92.07, Spanish 81.67, Kannada 74.67, English 60.21 and German 50.51. German's QR score was only 9.09.
  • Subtask B (Llama-3 8B): PR 95.00, CCNC 95.00 and QR 97.50, for an average of 95.83.
  • Stated limitations: The authors report semantic drift and over-neutralization in Tamil and Kannada, dependence on the size of the dataset, and subjectivity in LLM-as-a-Judge scoring.

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 6The 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 worksSocial Sciences · 1,954 worksGender Studies · 52 works

🧭 Topics 3Research subjects and languages the work is about.Multilingual & cross-lingual NLP · 882 worksBias, fairness & safety · 587 worksText generation · 214 works

Cite

Cite as Rajendran S, Ramkumar N and Malarselvi R. 2026. Igniters@LTEDI 2026: Multilingual Gender-Inclusive Language Generation with mT5 and Counter-Narrative Generation Using Llama-3. 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-1.18

@inproceedings{s2026igniters,
  title = {{Igniters@LTEDI 2026: Multilingual Gender-Inclusive Language Generation with mT5 and Counter-Narrative Generation Using Llama-3}},
  author = {Rajendran S, Ramkumar N and Malarselvi R},
  booktitle = {Proceedings of the Sixth Workshop on Language Technology for Equality, Diversity, Inclusion},
  year = {2026},
  doi = {10.18653/v1/2026.ltedi-1.18},
  url = {https://doi.org/10.18653/v1/2026.ltedi-1.18}
}

References

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

  1. Tolga Bolukbasi, Kai-Wei Chang, James Zou, Venkatesh Saligrama, and Adam Kalai. 2016. Man is to computer programmer as woman is to homemaker? debiasing word embeddings. In Advances in Neural Information Processing Systems.
  2. Bharathi Raja Chakravarthi, Shunmuga Priya Muthusamy Chinnan, Paul Buitelaar, Meghann L. Drury-Grogan, Miguel Ángel García Cumbreras, Salud María Jiménez Zafra, Thomas Mandl, Sylvia Jaki, Rahul Ponnusamy, Anand Kumar M, Dhanalakshmi V, Bharathi B, Premjith B, Senthil Kumar B, and Sathiyaraj T. 2026.… find
  3. Shunmuga Priya Muthusamy Chinnan, Meghann Drury-Grogan, and Bharathi Raja Chakravarthi. 2025. Gender inclusive language generation framework: A reasoning approach with retrieval augmented generation and chain-of-thought. Knowledge-Based Systems. find
  4. Tony Sun, Andrew Gaut, Shirlyn Tang, and Yuxin Huang. 2019. Mitigating gender bias in natural language processing. In ACL. find
  5. Eva Vanmassenhove, Chris Emmery, and Dimitar Shterionov. 2021. Neutral rewriter: A rule-based and neural approach to automatic rewriting into gender-neutral alternatives. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 8940–8948. find
  6. Tom Watson and 1 others. 2024. Fine-tuning with gender-inclusive language for bias mitigation in large language models. arXiv preprint arXiv:2407.04434. — [arXiv:2407.04434](https://arxiv.org/abs/2407.04434) find
  7. 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. Proceedings of NAACL-HLT.
  8. Jieyu Zhao, Tianlu Wang, Mark Yatskar, Vicente Ordonez, and Kai-Wei Chang. 2018. Gender bias in coreference resolution. In NAACL.
  9. Thomas Davidson, Dana Warmsley, Michael Macy, and Ingmar Weber. 2017. Automated hate speech detection and the problem of offensive language. In ICWSM.
  10. Emily Dinan, Stephen Roller, Kurt Shuster, Angela Fan, Michael Auli, and Jason Weston. 2019. Build it break it fix it for dialogue safety: Robustness from adversarial human attack. In EMNLP.
  11. Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aishwarya Rao, and 1 others. 2024. The llama 3 herd of models. arXiv preprint arXiv:2407.21783.
  12. Antigoni-Maria Founta, Constantinos Djouvas, and Despoina Chatzakou. 2018. Large scale crowdsourcing and characterization of twitter hate speech. In ICWSM. find
  13. Binny Mathew, Punyajoy Saha, Seid Muhie Yimam, Chris Biemann, Pawan Goyal, and Animesh Mukherjee. 2019. Thou shall not hate: Countering online hate speech. In ICWSM. find
  14. Shunmuga Priya Muthusamy Chinnan, Meghann Drury-Grogan, and Bharathi Raja Chakravarthi. 2025. Gender inclusive language generation framework: A reasoning approach with rag and cot. Knowledge-Based Systems, 328:114092. find
  15. Debora Nozza, Federico Bianchi, and Dirk Hovy. 2019. What the mask? making sense of gender bias in bert. In ACL Workshop on Gender Bias in NLP. find
  16. Andrea Piergentili, Beatrice Savoldi, Matteo Negri, and Luisa Bentivogli. 2025. Gender-neutral rewriting in italian: Models, approaches, and trade-offs. arXiv preprint arXiv:2509.13480. find

Updated

↑ back to top

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