---
title: "TriVector@DravidianLangTech 2026: Abusive Tamil Text"
description: "This paper describes the TriVector team's entry to the DravidianLangTech@ACL 2026 shared… 1.15 F1. 16 references, cited by 1 paper here."
authors: ["Oarisa Rebayet", "Tahmima Hoque Eid", "Fawzia Tabassum", "Hasan Murad"]
year: 2026
venue: "Proceedings of the Sixth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages"
source: "https://aclanthology.org/2026.dravidianlangtech-1.67/"
doi: "10.18653/v1/2026.dravidianlangtech-1.67"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/rebayet-2026-trivector-dravidianlangtech-2026-abusive-tamil/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/rebayet-2026-trivector-dravidianlangtech-2026-abusive-tamil/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/18/) / TriVector@DravidianLangTech 2026: Abusive Tamil Text Detection on Social Media Using Lexicon-Augmented Transformers

# TriVector@DravidianLangTech 2026: Abusive Tamil Text Detection on Social Media Using Lexicon-Augmented Transformers

**Authors:** [Oarisa Rebayet](https://md.research.elamur.ai/free/en/authors/oarisa-rebayet/), Tahmima Hoque Eid, Fawzia Tabassum and [Hasan Murad](https://md.research.elamur.ai/free/en/authors/hasan-murad/) · **Year:** 2026 · **Venue:** [Proceedings of the Sixth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages](https://md.research.elamur.ai/free/en/venues/dravidianlangtech/) ([Workshop on Speech, Vision, and Language Technologies for Dravidian Languages 2026](https://md.research.elamur.ai/free/en/venues/dravidianlangtech/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.dravidianlangtech-1.67/) · [PDF](https://aclanthology.org/2026.dravidianlangtech-1.67.pdf) · [DOI 10.18653/v1/2026.dravidianlangtech-1.67](https://doi.org/10.18653/v1/2026.dravidianlangtech-1.67) · [OpenAlex](https://openalex.org/W7166881072)

**Citations:** 0 citations (OpenAlex, 2026-10-14) · [1](https://md.research.elamur.ai/free/en/papers/rebayet-2026-trivector-dravidianlangtech-2026-abusive-tamil/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
This paper describes the TriVector team's entry to the DravidianLangTech@ACL 2026 shared task on detecting abusive Tamil comments aimed at women. Such social media text is informal, morphologically rich and often mixed with English. The authors ask whether simple statistical lexical cues still help once strong multilingual transformers are used.

They build an abusive lexicon from the roughly balanced shared-task corpus using PMI, give each entry a severity score, and derive nine per-comment features. These include lexicon hits, sentiment and negation counts, a sarcasm heuristic, and script ratios. The features are passed through a small MLP and concatenated with mean-pooled transformer embeddings before classification. Four backbones are tested at two training lengths, each with and without the lexicon branch.

The best result, 81.71% macro F1 and 15th place, came from XLM-RoBERTa-large without lexical features. The lexicon helped mid-sized models but hurt the large one. The study gives practical guidance on when lightweight feature engineering is still worthwhile in low-resource, code-mixed classification.

## Key points
- **Data:** Tamil comments labelled Abusive vs. Non-Abusive, split into 2,921 train, 731 validation and 913 test samples. Tokens are mostly Tamil, with averages of about 14 words per comment.
- **Lexicon:**
  - Candidate words had to appear in at least 3 abusive documents.
  - The 150 words with the highest PMI were kept.
  - PMI was rescaled from [−3, 3] to a 0–5 severity scale.
- **Fusion architecture:** The nine features are projected to 64 dimensions and combined with pooled embeddings in a classifier with a 128-unit hidden layer and dropout of 0.2. Training used 4 or 7 epochs, a learning rate of 2e-5 and a batch size of 16.
- **Models and results:**
  - The backbones were XLM-RoBERTa-base/large, MuRIL-base and DeOffXLMR-Tamil.
  - The largest lexicon gain was +2.81 F1, for XLM-R-base at 4 epochs (reaching 80.81%).
  - XLM-R-large lost 1.15 F1 at 4 epochs and 2.87 F1 at 7 epochs when the lexicon was added.
- **Limitations:** PMI can assign high scores to the names of frequently targeted public figures, introducing named-entity bias. The authors suggest filtering named entities and adding Tamil-specific morphological cues. Code is released on GitHub (OarisaR/TriVector-abusive-tamil-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.*

## Institutions

- [Chittagong University of Engineering & Technology](https://md.research.elamur.ai/free/en/universities/chittagong-university-of-engineering-technology/) — University, Chittagong, Bangladesh · 28 works · 29 citations

📚 Shelves (2): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works

## Cite

Cite as: Oarisa Rebayet, Tahmima Hoque Eid, Fawzia Tabassum and Hasan Murad. 2026. TriVector@DravidianLangTech 2026: Abusive Tamil Text Detection on Social Media Using Lexicon-Augmented Transformers. 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.

```bibtex
@inproceedings{rebayet2026trivector,
  title = {{TriVector@DravidianLangTech 2026: Abusive Tamil Text Detection on Social Media Using Lexicon-Augmented Transformers}},
  author = {Oarisa Rebayet, Tahmima Hoque Eid, Fawzia Tabassum 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-1.67},
  url = {https://doi.org/10.18653/v1/2026.dravidianlangtech-1.67}
}
```

- [← previous TrietNLP at ATE-IT: A Hybrid Pipeline for Italian Waste Management Terminology Analysis](https://md.research.elamur.ai/free/en/papers/triet-2026-trietnlp-ate-it-hybrid-pipeline/)
- [next → Truth or Mirage? Towards End-To-End Factuality Evaluation with LLM-O asis](https://md.research.elamur.ai/free/en/papers/scire-2026-truth-or-mirage-end-end/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/18/)

## References

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

1. Bharathi B and Josephine Varsha. 2022. SSNCSE NLP@TamilNLP-ACL2022: Transformer based approach for detection of abusive comment for Tamil language. In Proceedings of the Second Workshop on Speech and Language Technologies for Dravidian Languages, pages 158–164, Dublin, Ireland. Association for Compu… [find](https://scholar.google.com/scholar?q=Bharathi+B+and+Josephine+Varsha%2E+2022%2E+SSNCSE+NLP%40TamilNLP%2DACL2022%3A+Transformer+based+approach+for+detection+of+abusive+comment+for+Tamil+language%2E+In+Proceedings+of+the+Second+Workshop+on+Speech+and+)
2. Vimala Balakrishnan, Vithyatheri Govindan, and Kumanan N. Govaichelvan. 2023. Tamil offensive language detection: Supervised versus unsupervised learning approaches. ACM Trans. Asian Low-Resour. Lang. Inf. Process., 22(4). [find](https://scholar.google.com/scholar?q=Vimala+Balakrishnan%2C+Vithyatheri+Govindan%2C+and+Kumanan+N%2E+Govaichelvan%2E+2023%2E+Tamil+offensive+language+detection%3A+Supervised+versus+unsupervised+learning+approaches%2E+ACM+Trans%2E+Asian+Low%2DResour%2E+Lang%2E)
3. Bharathi Raja Chakravarthi, Ruba Priyadharshini, Navya Jose, Anand Kumar M, Thomas Mandl, Prasanna Kumar Kumaresan, Rahul Ponnusamy, Hariharan R L, John P. McCrae, and Elizabeth Sherly. 2021. [Findings of the shared task on offensive language identification in Tamil, Malayalam, and Kannada](https://md.research.elamur.ai/free/en/papers/chakravarthi-2021-findings-shared-task-offensive-language/). In Procee…
4. 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](https://md.research.elamur.ai/free/en/papers/conneau-2020-unsupervised-cross-lingual-representation-learning/). CoRR, abs/1911.02116.
5. Nikhil Ghanghor, Parameswari Krishnamurthy, Sajeetha Thavareesan, Ruba Priyadharshini, and Bharathi Raja Chakravarthi. 2021. IIITK@DravidianLangTech-EACL2021: Offensive language identification and meme classification in Tamil, Malayalam and Kannada. In Proceedings of the First Workshop on Speech and… [find](https://scholar.google.com/scholar?q=Nikhil+Ghanghor%2C+Parameswari+Krishnamurthy%2C+Sajeetha+Thavareesan%2C+Ruba+Priyadharshini%2C+and+Bharathi+Raja+Chakravarthi%2E+2021%2E+IIITK%40DravidianLangTech%2DEACL2021%3A+Offensive+language+identification+and+mem)
6. Hongyu Gong, Alberto Valido, Katherine M. Ingram, Giulia Fanti, Suma Bhat, and Dorothy L. Espelage. 2021. Abusive language detection in heterogeneous contexts: Dataset collection and the role of supervised attention. Proceedings of the AAAI Conference on Artificial Intelligence, 35(17):14804–14812. [find](https://scholar.google.com/scholar?q=Hongyu+Gong%2C+Alberto+Valido%2C+Katherine+M%2E+Ingram%2C+Giulia+Fanti%2C+Suma+Bhat%2C+and+Dorothy+L%2E+Espelage%2E+2021%2E+Abusive+language+detection+in+heterogeneous+contexts%3A+Dataset+collection+and+the+role+of+super)
7. Simran Khanuja, Diksha Bansal, Sarvesh Mehtani, Savya Khosla, Atreyee Dey, Balaji Gopalan, Dilip Kumar Margam, Pooja Aggarwal, Rajiv Teja Nagipogu, Shachi Dave, Shruti Gupta, Subhash Chandra Bose Gali, Vishnu K Subramanian, and Partha Pratim Talukdar. 2021. [Muril: Multilingual representations for in…](https://md.research.elamur.ai/free/en/papers/khanuja-2021-muril-multilingual-representations-indian-languages/)
8. Rohith Gowtham Kodali, Durga Prasad Manukonda, and Maharajan Pannakkaran. 2025. byte-SizedLLM@DravidianLangTech 2025: Abusive Tamil and Malayalam text targeting women on social media using XLM-RoBERTa and attention-BiLSTM. In Proceedings of the Fifth Workshop on Speech, Vision, and Language Technolo… [find](https://scholar.google.com/scholar?q=Rohith+Gowtham+Kodali%2C+Durga+Prasad+Manukonda%2C+and+Maharajan+Pannakkaran%2E+2025%2E+byte%2DSizedLLM%40DravidianLangTech+2025%3A+Abusive+Tamil+and+Malayalam+text+targeting+women+on+social+media+using+XLM%2DRoBERTa)
9. Ayush Maheshwari, Atul Kumar Singh, N J Karthika, Krishnakant Bhatt, Preethi Jyothi, and Ganesh Ramakrishnan. 2025. LexGen: Domain-aware multilingual lexicon generation. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 7364–737… [find](https://scholar.google.com/scholar?q=Ayush+Maheshwari%2C+Atul+Kumar+Singh%2C+N+J+Karthika%2C+Krishnakant+Bhatt%2C+Preethi+Jyothi%2C+and+Ganesh+Ramakrishnan%2E+2025%2E+LexGen%3A+Domain%2Daware+multilingual+lexicon+generation%2E+In+Proceedings+of+the+63rd+Ann)
10. MD.Mahadi Rahman, Mohammad Minhaj Uddin, and Mohammad Shamsul Arefin. 2025. CUET_Ignite@DravidianLangTech 2025: Detection of abusive comments in Tamil text using transformer models. In Proceedings of the Fifth Workshop on Speech, Vision, and Language Technologies for Dravidian Languages, pages 392–3… [find](https://scholar.google.com/scholar?q=MD%2EMahadi+Rahman%2C+Mohammad+Minhaj+Uddin%2C+and+Mohammad+Shamsul+Arefin%2E+2025%2E+CUET%5FIgnite%40DravidianLangTech+2025%3A+Detection+of+abusive+comments+in+Tamil+text+using+transformer+models%2E+In+Proceedings+of+)
11. Ratnavel Rajalakshmi, Ankita Duraphe, and Antonette Shibani. 2022. DLRG@DravidianLangTech-ACL2022: Abusive comment detection in Tamil using multilingual transformer models. In Proceedings of the Second Workshop on Speech and Language Technologies for Dravidian Languages, pages 207– 213, Dublin, Irel… [find](https://scholar.google.com/scholar?q=Ratnavel+Rajalakshmi%2C+Ankita+Duraphe%2C+and+Antonette+Shibani%2E+2022%2E+DLRG%40DravidianLangTech%2DACL2022%3A+Abusive+comment+detection+in+Tamil+using+multilingual+transformer+models%2E+In+Proceedings+of+the+Secon)
12. Tamil Text Ratnavel Rajalakshmi, Srivarshan Selvaraj, Faerie Mat-tins R., Pavitra Vasudevan, and Anand Kumar M. HOTTEST: Hate and offensive content 2023. identification in Tamil using transformers and enhanced STemming. Computer Speech & Language, 78:101464. [find](https://scholar.google.com/scholar?q=Tamil+Text+Ratnavel+Rajalakshmi%2C+Srivarshan+Selvaraj%2C+Faerie+Mat%2Dtins+R%2E%2C+Pavitra+Vasudevan%2C+and+Anand+Kumar+M%2E+HOTTEST%3A+Hate+and+offensive+content+2023%2E+identification+in+Tamil+using+transformers+and)
13. Saranya Rajiakodi, Shunmuga Priya Muthusamy Chinnan, Rajalakshmi R., Kathiravan Pannerselvam, Bhuvaneswari Sivagnanam, Jananayagan V, Charmathi Rajkumar, R Ramesh Kannan, and Bharathi Raja Chakravarthi. 2026. [From Comments to Harm: A Findings Report on Abusive Tamil Text Targeting Women on Social Me…](https://md.research.elamur.ai/free/en/papers/sivagnanam-2026-comments-harm-findings-report-abusive/)
14. Varun Balaji S, Bojja Revanth Reddy, Vyshnavi Reddy Battula, Suraj Nagunuri, and Core-Balasubramanian Palani. 2025. Four_IIITK@DravidianLangTech 2025: Abusive content detection against women using machine learning and deep learning models. In Proceedings of the Fifth Workshop on Speech, Vision, and… [find](https://scholar.google.com/scholar?q=Varun+Balaji+S%2C+Bojja+Revanth+Reddy%2C+Vyshnavi+Reddy+Battula%2C+Suraj+Nagunuri%2C+and+Core%2DBalasubramanian+Palani%2E+2025%2E+Four%5FIIITK%40DravidianLangTech+2025%3A+Abusive+content+detection+against+women+using+mac)
15. Debjoy Saha, Naman Paharia, Debajit Chakraborty, Punyajoy Saha, and Animesh Mukherjee. 2021. Hate-alert@DravidianLangTech-EACL2021: Ensembling strategies for transformer-based offensive language detection. In Proceedings of the First Workshop on Speech and Language Technologies for Dravidian Languag… [find](https://scholar.google.com/scholar?q=Debjoy+Saha%2C+Naman+Paharia%2C+Debajit+Chakraborty%2C+Punyajoy+Saha%2C+and+Animesh+Mukherjee%2E+2021%2E+Hate%2Dalert%40DravidianLangTech%2DEACL2021%3A+Ensembling+strategies+for+transformer%2Dbased+offensive+language+detec)
16. Advaitha Vetagiri, Gyandeep Kalita, Eisha Halder, Chetna Taparia, Partha Pakray, and Riyanka Manna. 2024. Breaking the silence detecting and mitigating gendered abuse in hindi, tamil, and indian english online spaces. ArXiv, abs/2404.02013. — [arXiv:2404.02013](https://arxiv.org/abs/2404.02013) [find](https://scholar.google.com/scholar?q=Advaitha+Vetagiri%2C+Gyandeep+Kalita%2C+Eisha+Halder%2C+Chetna+Taparia%2C+Partha+Pakray%2C+and+Riyanka+Manna%2E+2024%2E+Breaking+the+silence+detecting+and+mitigating+gendered+abuse+in+hindi%2C+tamil%2C+and+indian+engli)

*Updated 2 November 2026*