Elamur Research⌕
☰
home / papers / 2026 / CUET_SYNTHETICA@DravidianLangTech 2026: Multilingual Transformer Based Hope Speech Detection for Coarse and Fine-Grained Classification in Tulu

CUET_SYNTHETICA@DravidianLangTech 2026: Multilingual Transformer Based Hope Speech Detection for Coarse and Fine-Grained Classification in Tulu

Paper · 2026 · Sumaiya Zaman, Miftahul Jannat Rishta, Shiti Chowdhury et al.

Summary

This shared-task paper tackles hope speech detection in code-mixed Tulu. Tulu is a low-resource Dravidian language of coastal Karnataka, and YouTube users often write it mixed with English and Kannada. The authors entered both subtasks of the DravidianLangTech@ACL 2026 challenge:

  • Task 1: coarse four-way labelling of hope tone (Encouraging, Discouraging, Uninvolved, Blended).
  • Task 2: fine-grained five-way labelling of hope type (Optimistic, Realistic, Inspiring, Fading Hope, Hopelessness).

They fine-tuned three multilingual encoders, XLM-RoBERTa-base, MuRIL and mBERT. Preprocessing was light: text cleanup and fixed-length inputs of about 128 tokens. Task 2 also used class weighting. For Task 1, the submission averages softmax probabilities from all three models (soft voting). For Task 2, a single XLM-RoBERTa-base model was submitted.

Both systems ranked first, with test macro F1 of 0.58 and 0.42. These modest absolute scores show the problem remains hard, especially for ambiguous categories. Still, the released code gives a reproducible baseline for analysing positive content in an under-served code-mixed language.

Key points

  • Data: YouTube comments from the shared task. Task 1 has 8,401 items (5,991 train / 1,284 dev / 1,126 test). Task 2 has 4,550 items (3,185 / 682 / 683).
  • Official test results: The Task 1 three-model ensemble reached macro F1 0.58 (accuracy 0.67). Task 2 XLM-RoBERTa-base reached macro F1 0.42 (accuracy 0.50). Both placed 1st.
  • Training setup: AdamW, batch size 16, learning rate 2e-5 (1.5e-5 for XLM-R in Task 2), 4–10 epochs. Training ran on a single NVIDIA T4 on Kaggle and took about 3 hours for the full ensemble. Code: github.com/sumaiyaa110/Hope-Speech-Detection.
  • Error analysis: Blended Hope was the most confused Task 1 class. Realistic Hope was the hardest Task 2 class. Errors are attributed to class imbalance and inconsistent tokenization of code-mixed text.
  • Caveat: The reported dev scores do not fully support the model choices.
    • In Task 1, XLM-R alone had a higher dev F1 (0.6509) than the ensemble (0.6179).
    • In Task 2, mBERT (0.4414) outscored XLM-R (0.4218) on dev.

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

🎓Chittagong University of Engineering & TechnologyChittagong · Bangladesh28 works · 29 citations

 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 worksPsychology · 1,649 worksSocial Psychology · 330 works

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

Cite

Cite as Sumaiya Zaman, Miftahul Jannat Rishta, Shiti Chowdhury and Hasan Murad. 2026. CUET_SYNTHETICA@DravidianLangTech 2026: Multilingual Transformer Based Hope Speech Detection for Coarse and Fine-Grained Classification in Tulu. Proceedings of the Sixth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages.

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

@inproceedings{zaman2026cuet,
  title = {{CUET\_SYNTHETICA@DravidianLangTech 2026: Multilingual Transformer Based Hope Speech Detection for Coarse and Fine-Grained Classification in Tulu}},
  author = {Sumaiya Zaman, Miftahul Jannat Rishta, Shiti Chowdhury and Hasan Murad},
  booktitle = {Proceedings of the Sixth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages},
  year = {2026},
  doi = {10.18653/v1/2026.dravidianlangtech-1.30},
  url = {https://doi.org/10.18653/v1/2026.dravidianlangtech-1.30}
}

References

16 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Pranjal Aggarwal, Pasupuleti Chandana, Jagrut Nemade, Shubham Sharma, Sunil Saumya, and Shankar Biradar. 2022. Hope speech detection on social media platforms. arXiv preprint arXiv:2212.07424. — [arXiv:2212.07424](https://arxiv.org/abs/2212.07424) find
  2. Fazlourrahman Balouchzahi, Grigori Sidorov, and Alexander Gelbukh. 2023. Polyhope: Two-level hope speech detection from tweets. Expert Systems with Applications, 222:120078. find
  3. Bharathi Raja Chakravarthi. 2020. HopeEDI: A multilingual hope speech detection dataset for equality, diversity, and inclusion. In Proceedings of the Third Workshop on Computational Modeling of People’s Opinions, Personality, and Emotion’s in Social Media, pages 41– 53, Barcelona, Spain (Online). As…
  4. Bharathi Raja Chakravarthi. 2022. Multilingual hope speech detection in english and dravidian languages. BMC Bioinformatics, 23(303). find
  5. Bharathi Raja Chakravarthi and Vigneshwaran Muralidaran. 2021. Findings of the shared task on hope speech detection for equality, diversity, and inclusion. In Proceedings of the First Workshop on Language Technology for Equality, Diversity and Inclusion, pages 61–72, Kyiv. Association for Computatio…
  6. Bharathi Raja Chakravarthi, Vigneshwaran Muralidaran, Ruba Priyadharshini, Subalalitha Chinnaudayar Navaneethakrishnan, John Philip McCrae, Miguel Ángel García-Cumbreras, Salud María Jiménez-Zafra, Rafael Valencia-García, Prasanna Kumar Kumaresan, Rahul Ponnusamy, Daniel García-Baena, and José Anton… find
  7. Multilingual hope speech detection: A robust framework using transfer learning of fine-tuning roberta model. Journal of King Saud University – Computer and Information Sciences, 35(8):101736. find
  8. Mohammad Zia Ur Rehman, Devraj Raghuvanshi, Harshit Pachar, Chandravardhan Singh Raghaw, and Nagendra Kumar. 2025. Hierarchical attention-enhanced contextual capsulenet for multilingual hope speech detection. Expert Systems with Applications, 268:126285. find
  9. Grigori Sidorov, Fazlourrahman Balouchzahi, Luis Ramos, Helena Gómez-Adorno, and Alexander Gelbukh. 2025. Multilingual identification of nuanced dimensions of hope speech in social media. Journal of Biomedical Informatics, 150:104512. find
  10. Arunima Sundar, Akshay Ramakrishnan, Avantika Balaji, and Thenmozhi Durairaj. 2021. Hope speech detection for dravidian languages using cross-lingual embeddings with stacked encoder architecture. SN Computer Science, 2(1):Article 27. find
  11. Durairaj Thenmozhi, Rathnakar Shetty P, Parameshwar R. Hegde, Anusha M D, Raksha Adyanthaya, Mohammed Fadhel Aljunid, Prasanna Kumar Kumaresan, and Bharathi Raja Chakravarthi. 2026. Findings of the Shared Task on Hope Speech Detection in Tulu. In Proceedings of the Sixth Workshop on Speech, Vision,… find
  12. Alexis Conneau, Guillaume Lample, Sebastian Ruder, et al. 2019. Unsupervised cross-lingual representation learning at scale. arXiv preprint arXiv:1911.02116.
  13. Mithun Das, Shubhankar Barman, and Subhadeep Chatterjee. 2023. hate-alert@lt-edi-2023: Hope speech detection using transformer-based models. In Proceedings of the Third Workshop on Language Technology for Equality, Diversity and Inclusion, pages 250–256, Varna, Bulgaria. INCOMA Ltd., Shoumen, Bulgar… find
  14. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805.
  15. Sanchit Khanuja et al. 2021. Muril: Multilingual representations for indian languages. arXiv preprint arXiv:2103.10730.
  16. Muhammad Shahid Iqbal Malik, Anna Nazarova, Mona Mamdouh Jamjoom, and Dmitry I. Ignatov. 2023. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/zaman-2026-cuet-synthetica-dravidianlangtech-2026-multilingual/