Elamur Research⌕
☰
home / papers / 2026 / CAI@LTEDI 2026: Multilingual Gender Inclusive Language Generation using Instruction-Guided mT5 Transformer Model

CAI@LTEDI 2026: Multilingual Gender Inclusive Language Generation using Instruction-Guided mT5 Transformer Model

Paper · 2026 · Aiswariya p Nair, Sree S Bhagya and Chinnu Jacob

Summary

The paper addresses rewriting gender-biased or gender-marked sentences into neutral versions that keep the original meaning. The work is part of the LT-EDI 2026 shared task, which covers English, German, Spanish, Tamil, and Kannada. Team CAI merged the organisers' parallel datasets into one corpus and fine-tuned a single mT5-small sequence-to-sequence model on it, marking every input with a "rewrite:" task prefix. During inference, they used beam search with repetition constraints, because greedy decoding had produced truncated, repetitive, or language-mixed output, especially for Tamil and Kannada. Evaluation combined a fairness index (GIFI), embedding-based semantic similarity, and a length term. In the official ranking, the system placed 4th for German, Spanish, and Kannada, 5th for Tamil, and 7th for English. Error analysis found truncation, repetition, over-neutralization, and code-mixing. The study matters as a low-cost baseline: one compact model can attempt inclusive rewriting across typologically different languages, including Dravidian ones, though its limits are clear.

Key points

  • Training data (Subtask A) consists of sentence pairs: 1074 each for English, Tamil, and Kannada, 1002 for German, and only 200 for Spanish.
  • The fine-tuning setup was mT5-small with AdamW, learning rate 3e-5, batch size 8, 5 epochs, max sequence length 128, and beam width 4, using HuggingFace Transformers. mT5-base was not tried because of compute limits.
  • The metrics were defined as follows:
    • GIFI = 0.6·bias removal + 0.4·inclusive term usage.
    • Overall score = 0.5·GIFI + 0.4·similarity + 0.1·length.
  • Official average scores per language were Kannada 64.00, English 56.88, Tamil 50.68, Spanish 45.83, and German 14.71. Two languages scored 0 on the QR metric: Kannada and Spanish scored 5.00 and German scored 0.00 on that column.
  • The authors suggest future work on larger datasets, better decoding, morphology-aware modelling, and discourse context, since the current method is sentence-level only.

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

Also on the byline: TKM College of Engineering.

 Fields, topics and concepts

📚 Shelves 6The 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 worksSocial Sciences · 1,772 worksGender Studies · 52 works

🧭 Topics 2Research subjects and languages the work is about.Multilingual & cross-lingual NLP · 828 worksText generation · 202 works

Cite

Cite as Aiswariya p Nair, Sree S Bhagya and Chinnu Jacob. 2026. CAI@LTEDI 2026: Multilingual Gender Inclusive Language Generation using Instruction-Guided mT5 Transformer Model. 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-1.14

@inproceedings{nair2026cai,
  title = {{CAI@LTEDI 2026: Multilingual Gender Inclusive Language Generation using Instruction-Guided mT5 Transformer Model}},
  author = {Aiswariya p Nair, Sree S Bhagya and Chinnu Jacob},
  booktitle = {Proceedings of the Sixth Workshop on Language Technology for Equality, Diversity, Inclusion},
  year = {2026},
  doi = {10.18653/v1/2026.ltedi-1.14},
  url = {https://doi.org/10.18653/v1/2026.ltedi-1.14}
}

References

14 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Faezeh Torabi Asr, Maite Taboada, and Alexandra J. Y. Cheng. 2021. The gender gap tracker: Using natural language processing to measure gender bias in media. PLOS ONE. find
  2. Marion Bartl, Malvina Nissim, and Albert Gatt. 2020. Unmasking contextual stereotypes: Measuring and mitigating bert’s gender bias. In ACL Workshop on Gender Bias in NLP.
  3. 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.
  4. Bharathi Raja Chakravarthi, Shunmuga Priya Muthusamy Chinnan, Paul Buitelaar, and Meghann Drury-Grogan. 2026a. Insights from multilingual gender inclusive language generation shared task. In Proceedings of the Sixth Workshop on Language Technology for Equality, Diversity, Inclusion. find
  5. 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. 2026b… find
  6. Emily Dinan, Angela Fan, Adina Williams, Jack Urbanek, Douwe Kiela, and Jason Weston. 2020. Queens are powerful too: Mitigating gender bias in dialogue generation. In Proceedings of EMNLP.
  7. Enzo Doyen and Amalia Todirascu. 2025. Genre: A french gender-neutral rewriting system using collective nouns. In Findings of the Association for Computational Linguistics: ACL 2025, pages 7889–7909. find
  8. Dirk Hovy and Shannon L. Spruit. 2021. The social impact of natural language processing. Computational Linguistics. find
  9. Paul Lerner and Cyril Grouin. 2024. Inclure: a dataset and toolkit for inclusive french translation. In Proceedings of the 17th Workshop on Building and Using Comparable Corpora (BUCC)@ LREC-COLING 2024, pages 59–68. find
  10. Haochen Liu, Wentao Wang, Yiqi Wang, Hui Liu, Zitao Liu, and Jiliang Tang. 2020. Mitigating gender bias for neural dialogue generation with adversarial learning. In Proceedings of EMNLP. find
  11. Emily Sheng, Kai-Wei Chang, Premkumar Natarajan, and Nanyun Peng. 2019. The woman worked as a babysitter: On biases in language generation. In Proceedings of EMNLP.
  12. Karolina Stańczak and Isabelle Augenstein. 2021. A survey on gender bias in natural language processing. Journal of the ACM. find
  13. Leonor Veloso, Luisa Coheur, and Rui Ribeiro. 2023. A rewriting approach for gender inclusivity in portuguese. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 8747–8759. find
  14. Shunmuga Priya Muthusamy Chinnan, Meghann Drury-Grogan, and Bharathi Raja Chakravarthi. 2025. Gender inclusive language generation framework: A rea-https://github.com/123veno/LTEDI_ 2026-GENDER-INCLUSIVE-LANGUAGE-GENERATION find

Updated

↑ back to top

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