Elamur Research⌕
☰
home / papers / 2026 / PrimeLine@DravidianLangTech 2026: Hope Speech Detection in Tulu Using XLM-RoBERTa for Coarse and Fine-Grained Classification

PrimeLine@DravidianLangTech 2026: Hope Speech Detection in Tulu Using XLM-RoBERTa for Coarse and Fine-Grained Classification

Paper · 2026 · Rithikaa V, S.Sumathi, Sanjay Krishnan K et al.

Summary

The paper addresses hope speech classification for Tulu, a Dravidian language of coastal Karnataka and Kerala with roughly 4–5 million speakers but almost no NLP resources. Tulu social media comments freely combine Tulu, Kannada script, English and emoji, which makes automatic labeling hard. Within the DravidianLangTech@ACL 2026 shared task, the authors fine-tune xlm-roberta-base with a linear head on the [CLS] representation for two tracks: four coarse hope categories and five fine-grained ones. Raw text goes directly into the SentencePiece tokenizer. They skip transliteration, language identification and script normalization so that code-switching signals stay intact. Training is short (2 epochs on a Colab T4) and does not correct for class imbalance. Official Codabench results are Macro F1 0.34 on Track 1 and 0.19 on Track 2. Most errors come from the dominant Uninvolved class absorbing minority labels and from fine-grained categories that are nearly indistinguishable. The main value of the work is that it provides the first transformer baseline for this task in Tulu, with substantial room for improvement.

Key points

  • Data: Track 1 has 5,991 training comments labeled Encouraging, Discouraging, Blended or Uninvolved. The set is imbalanced: Uninvolved makes up 41.6% and Discouraging Hope only 11.9%. Track 2 has 3,185 comments labeled Inspiring, Hopelessness, Realistic, Optimistic or Fading Hope, with Fading Hope as the smallest class (236 samples, 7.4%).
  • Setup: xlm-roberta-base (~270M parameters), max length 128, batch size 8, AdamW (lr 2e-5, weight decay 0.01), 6% linear warmup. The checkpoint was selected by dev-set loss, and no ensembling was used.
  • Results (Codabench, macro-averaged):
    • Track 1: accuracy 0.60, P 0.30, R 0.41, F1 0.34.
    • Track 2: accuracy 0.24, P/R/F1 0.19 each. Track 2 accuracy is below the majority-class share of 35.4% in the training data.
  • Proposed next steps:
    • Weighted loss or back-translation augmentation from Kannada.
    • Longer training with early stopping on macro F1.
    • Dravidian-specific or larger multilingual models.
    • Contrastive objectives for separating similar categories.
  • Code: github.com/Rithikaa7/PrimeLine-Hope_Speech_Detection

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 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

Cite

Cite as Rithikaa V, S.Sumathi, Sanjay Krishnan K and Nithya Varshini C N R. 2026. PrimeLine@DravidianLangTech 2026: Hope Speech Detection in Tulu Using XLM-RoBERTa for Coarse and Fine-Grained Classification. 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-1.52

@inproceedings{v2026primeline,
  title = {{PrimeLine@DravidianLangTech 2026: Hope Speech Detection in Tulu Using XLM-RoBERTa for Coarse and Fine-Grained Classification}},
  author = {Rithikaa V, S.Sumathi, Sanjay Krishnan K and Nithya Varshini C N R},
  booktitle = {Proceedings of the Sixth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages},
  year = {2026},
  doi = {10.18653/v1/2026.dravidianlangtech-1.52},
  url = {https://doi.org/10.18653/v1/2026.dravidianlangtech-1.52}
}

References

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

  1. Fazlourrahman Balouchzahi, Aparna B K, and H. L. Shashirekha. MUCS@LT-EDI-EACL2021: CoHope – hope speech detection for equality, diversity, and inclusion in code-mixed texts. In Proceedings of the First Workshop on Language Technology for Equality, Diversity and Inclusion, pages 180–187, 2021. URL h… find
  2. Bharathi Raja Chakravarthi. 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, 2020. URL https://aclanthology.org/… ↗
  3. Bharathi Raja Chakravarthi. Multilingual hope speech detection in English and Dravidian languages. International Journal of Data Science and Analytics, 14 (4):389–406, 2022. URL https://doi.org/10. 1007/s41060-022-00324-5. find
  4. Bharathi Raja Chakravarthi and Vigneshwaran Muralidaran. 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, 2021. URL https://aclanthology.org/ 2… ↗
  5. Bharathi Raja Chakravarthi et al. Overview of the shared task on hope speech detection for equality, diversity, and inclusion. In Proceedings of the Second Workshop on Language Technology for Equality, Diversity and Inclusion, pages 378–388, 2022. URL https://aclanthology.org/2022.ltedi-1.53. ↗
  6. Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. Unsupervised cross-lingual representation learning at scale. In Proceedings of the 58th Annual Meeting of the Association for Comp… find
  7. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. BERT: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of NAACL-HLT 2019, pages 4171– 4186, 2019. URL https://aclanthology.org/ N19-1423. [10] Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Ma… ↗
  8. Toqeer Ehsan, Amina Tehseen, Kengatharaiyer Sarveswaran, and Amjad Ali. Al-phaBrains@DravidianLangTech: Sentiment analysis of code-mixed Tamil and Tulu using ELMo embeddings. In Proceedings of the Third Workshop on Speech and Language Technologies for Dravidian Languages, pages 152–159, 2023. URL ht… find
  9. Bo Huang and Yang Bai. TEAM HUB@LT-EDI-EACL2021: Hope speech detection based on pre-trained language model. In Proceedings of the First Workshop on Language Technology for Equality, Diversity and Inclusion, pages 122–127, 2021. URL https://aclanthology.org/2021.ltedi-1.17. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/v-2026-primeline-dravidianlangtech-2026-hope-speech/