Elamur Research⌕
☰
home / papers / 2026 / Team Oryu@DravidianLangTech 2026: A Multilingual Transformer Approach for Hope Speech Detection in Code-Mixed Tulu

Team Oryu@DravidianLangTech 2026: A Multilingual Transformer Approach for Hope Speech Detection in Code-Mixed Tulu

Paper · 2026 · Joyeta Barua Moni, Noore Tamanna Orny, Md. Abtahee Kabir et al.

Summary

This paper describes Team Oryu's submission to Task 1 of the DravidianLangTech 2026 shared task. The task asks systems to sort code-mixed Tulu YouTube comments into four coarse hope-tone categories, with each comment tagged as either Encouraging or Discouraging, or else as Blended Tone or Uninvolved. The comments mix Tulu with English and Kannada and are mostly in Roman script. Tulu is an under-resourced Dravidian language, and the organizers present this task as its first hope speech benchmark.

The authors fine-tune XLM-RoBERTa-base with a classification head. Preprocessing is light: they strip URLs and extra whitespace and shorten elongated character runs, but keep emojis as emotional signals. To offset the dominance of the Uninvolved class, they weight the cross-entropy loss by class frequency. They also add training variants by randomly deleting tokens.

On the hidden test set the system reached a Macro F1 of 63% and placed 3rd. Blended Tone and sarcastic or mixed-emotion comments remained the hardest cases. The work gives a simple, reproducible baseline, with code on GitHub, for affect-oriented classification in a code-mixed low-resource language.

Key points

  • Dataset (Thenmozhi et al., 2026): 5,991 train, 1,284 dev and 1,126 test comments. Uninvolved is the largest class (2,490 in train) and Discouraging the smallest (711).
  • Training setup: XLM-RoBERTa-base trained with AdamW in the Hugging Face Trainer.
    • Learning rate 2 × 10⁻⁵, batch size 16, 3 epochs, sequences of 128 tokens.
    • Random token deletion with probability 0.1.
  • Test results: Accuracy 67%, Macro Precision 64%, Macro Recall 65%, Macro F1 63% (3rd place).
  • Ablation (Macro F1):
    • Plain baseline: 0.5554.
    • With WeightedSampler: 0.6220.
    • Full system: 0.6276.
    • Most of the gain comes from class balancing; augmentation adds little.
  • Caveats to check before citing:
    • The paper says the top system scored 58% Macro F1, which conflicts with a 3rd-place finish at 63%.
    • The ablation table names a WeightedSampler, while the method section describes class-weighted loss.

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 · Bangladesh25 works · 29 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksPsychology · 1,552 worksSocial Psychology · 306 works

🧭 Topics 1Research subjects and languages the work is about.Multilingual & cross-lingual NLP · 828 works

Cite

Cite as Joyeta Barua Moni, Noore Tamanna Orny, Md. Abtahee Kabir and Hasan Murad. 2026. Team Oryu@DravidianLangTech 2026: A Multilingual Transformer Approach for Hope Speech Detection in Code-Mixed 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-1.64

@inproceedings{moni2026team,
  title = {{Team Oryu@DravidianLangTech 2026: A Multilingual Transformer Approach for Hope Speech Detection in Code-Mixed Tulu}},
  author = {Joyeta Barua Moni, Noore Tamanna Orny, Md. Abtahee Kabir 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-1.64},
  url = {https://doi.org/10.18653/v1/2026.dravidianlangtech-1.64}
}

References

10 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Ahmed Abdullah, Sana Fatima, and Haroon Mahmood. 2025. Ghalib: A multilingual framework for hope speech detection in low-resource languages. arXiv preprint arXiv:2512.22705. — [arXiv:2512.22705](https://arxiv.org/abs/2512.22705) find
  2. Muhammad Ahmad, Muhammad Waqas, Ameer Hamza, Ildar Batyrshin, and Grigori Sidorov. 2025. Hope speech detection in code-mixed roman urdu tweets: A positive turn in natural language processing. arXiv preprint arXiv:2506.21583. — [arXiv:2506.21583](https://arxiv.org/abs/2506.21583) find
  3. Bharathi Raja Chakravarthi. 2022a. Hope speech detection in youtube comments. Social Network Analysis and Mining, 12(1):75. find
  4. Bharathi Raja Chakravarthi. 2022b. Multilingual hope speech detection in english and dravidian languages. International Journal of Data Science and Analytics, 14(4):389–406. find
  5. Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2019. Unsupervised cross-lingual representation learning at scale. CoRR, abs/1911.02116. find
  6. Asha Hegde, Mudoor Devadas Anusha, Sharal Coelho, Hosahalli Lakshmaiah Shashirekha, and Bharathi Raja Chakravarthi. 2022. Corpus creation for sentiment analysis in code-mixed tulu text. In Proceedings of the 1st Annual Meeting of the ELRA/ISCA Special Interest Group on Under-Resourced Languages, pag… find
  7. Asha Hegde, Bharathi Raja Chakravarthi, Hosahalli Lakshmaiah Shashirekha, Rahul Ponnusamy, Subalalitha Cn, Lavanya SK, Martha Karunakar, Shreya Shreeram, Sarah Aymen, et al. 2023. Findings of the shared task on sentiment analysis in tamil and tulu code-mixed text. In Proceedings of the Third Worksho… find
  8. Anqi Mao, Mehryar Mohri, and Yutao Zhong. 2023. Cross-entropy loss functions: Theoretical analysis and applications. Preprint, arXiv:2304.07288. — [arXiv:2304.07288](https://arxiv.org/abs/2304.07288) find
  9. Ankit Kumar Mishra, Sunil Saumya, and Abhinav Kumar. 2021. Sentiment analysis of dravidian-codemix language. In FIRE (Working Notes), pages 1011– 1019. find
  10. 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

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/moni-2026-team-oryu-dravidianlangtech-2026-multilingual/