---
title: "TDMulti: A Tunisian Dialect-Modern Standard — Torjmen 2026"
description: "Tunisian dialect (TD) is the main language of Tunisian social media, yet NLP resources for it are scarce."
authors: ["Roua Torjmen", "Kais HADDAR"]
year: 2026
venue: "Proceedings of the Fifteenth Language Resources and Evaluation Conference"
source: "https://aclanthology.org/2026.lrec-1.254/"
doi: "10.63317/5aev569ryz8n"
licence: "CC BY-NC 4.0 (ELRA, LREC proceedings)"
licence_url: "https://creativecommons.org/licenses/by-nc/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/papers/torjmen-2026-tdmulti-tunisian-dialect-modern-standard/"
---
[Elamur Research](https://md.research.elamur.ai/) · [this page in HTML](https://research.elamur.ai/free/papers/torjmen-2026-tdmulti-tunisian-dialect-modern-standard/)

[home](https://md.research.elamur.ai/) / [papers](https://md.research.elamur.ai/free/papers/) / [2026](https://md.research.elamur.ai/free/papers/2026/page/4/) / TDMulti: A Tunisian Dialect-Modern Standard Arabic Multitask Corpus with a Context-Aware Cross-Attention BERT Model

# TDMulti: A Tunisian Dialect-Modern Standard Arabic Multitask Corpus with a Context-Aware Cross-Attention BERT Model

**Authors:** Roua Torjmen, Kais HADDAR · **Year:** 2026 · **Venue:** Proceedings of the Fifteenth Language Resources and Evaluation Conference · **Licence:** [CC BY-NC 4.0 (ELRA, LREC proceedings)](https://creativecommons.org/licenses/by-nc/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.lrec-1.254/) · [PDF](https://aclanthology.org/2026.lrec-1.254.pdf) · [DOI 10.63317/5aev569ryz8n](https://doi.org/10.63317/5aev569ryz8n) · [OpenAlex](https://openalex.org/W7160061002)

**Citations:** 0 citations (openalex-snapshot · 2026-10-14)

## Summary
Tunisian dialect (TD) is the main language of Tunisian social media, yet NLP resources for it are scarce. Existing TD datasets usually cover only one task, such as hate speech or sarcasm. They lack a Modern Standard Arabic (MSA) counterpart and are often not public. The authors address this with TDMulti, a corpus of 3,100 comments gathered from Tunisian Facebook pages, YouTube channels and forums between 2023 and 2025. Each comment is manually paired with an MSA equivalent and labelled for four related tasks: hate speech, sentiment polarity, sarcasm and topic category.

On top of this resource they build BERT-based multitask classifiers in three versions:
- a plain multitask baseline,
- a model applying cross-attention between TD and MSA representations,
- a context-aware cross-attention variant with masking specific to each task.

In experiments with MARBERT and AraBERT, the context-aware model on MARBERT performs best, with the clearest gains on sarcasm and hate speech. The openly released corpus and code give a first benchmark for jointly modelling pragmatic and social phenomena in this low-resource Arabic dialect.

## Key points
- TDMulti contains 3,100 TD social media comments, each manually aligned with MSA. It carries 12,400 labels in total, one per task for each of the four tasks.
- It is presented as the first public multitask TD–MSA resource. Earlier corpora such as T-HSAB, HateTune, Bully.tn, TEET!, PoliTun and the Mekki et al. sarcasm set are single-purpose or not openly available.
- Three architectures are compared: a baseline multitask classifier, TD–MSA cross-attention, and context-aware cross-attention with task-specific masking.
- The pre-trained encoders are MARBERT and AraBERT. MARBERT with context-aware cross-attention gives the strongest results, especially for sarcasm and hate speech detection.
- Comments were anonymized before release. The data and code are published at github.com/rouatorjmen1/MultiTD-Multitask-BERT. The authors acknowledge that topic-balanced sampling may not mirror the natural diversity of Tunisian online discourse.

*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: [Linguistics & languages](https://md.research.elamur.ai/free/shelves/linguistics/)

## Cite

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{torjmen2026tdmulti,
  title = {{TDMulti: A Tunisian Dialect-Modern Standard Arabic Multitask Corpus with a Context-Aware Cross-Attention BERT Model}},
  author = {Torjmen, Roua and HADDAR, Kais},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation Conference},
  year = {2026},
  doi = {10.63317/5aev569ryz8n},
  url = {https://doi.org/10.63317/5aev569ryz8n}
}
```

- [← previous Tao–Filipino Neural Machine Translation: Strategies for Ultra–Low-Resource Settings](https://md.research.elamur.ai/free/papers/macayan-2026-tao-filipino-neural-machine-translation/)
- [next → Teaching linguistic prompt control for LLM based translation: A classroom approach to developing critical and responsible AI literacy](https://md.research.elamur.ai/free/papers/menzel-2026-teaching-linguistic-prompt-control-llm/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/papers/2026/page/4/)

## References

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

1. Abbes, M., Kechaou, Z., & Alimi, A. M. (2023). Deep learning approach for Tunisian hate speech detection on Facebook. In Proceedings of the IEEE Symposium on Computers and Communications (ISCC 2023), pp. 739-744, Tunis, Tunisia, July. IEEE. [find](https://scholar.google.com/scholar?q=Abbes%2C+M%2E%2C+Kechaou%2C+Z%2E%2C+%26+Alimi%2C+A%2E+M%2E+%282023%29%2E+Deep+learning+approach+for+Tunisian+hate+speech+detection+on+Facebook%2E+In+Proceedings+of+the+IEEE+Symposium+on+Computers+and+Communications+%28ISCC+2023%29%2C+)
2. Abdaoui, A., Berrimi, M., Oussalah, M., & Moussaoui, A. (2021). DziriBERT: A pre-trained language model for the Algerian dialect. arXiv preprint arXiv:2109.12346. [↗](https://arxiv.org/abs/2109.12346)
3. Abdul-Mageed, M., Elmadany, A., & Nagoudi, E. M. B. (2021). ARBERT & MARBERT: Deep bidirectional transformers for Arabic. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (ACL/IJCNLP 2021), pp. 7088-7105, Online, August. ACL. [↗](https://doi.org/10.48550/arxiv.2101.01785) [our card](https://md.research.elamur.ai/free/papers/abdul-2021-mageed-arbert-marbert-deep-bidirectional-transformers/)
4. Antoun, W., Baly, F., & Hajj, H. M. (2020). AraBERT: Transformer-based model for Arabic language understanding. arXiv preprint arXiv:2003.00104. [↗](https://arxiv.org/abs/2003.00104) [our card](https://md.research.elamur.ai/free/papers/antoun-2020-arabert-transformer-based-model-arabic/)
5. Baazaoui, H., Cheniti-Belcadhi, L., & Zrigui, M. (2025). Hate speech detection in Tunisian dialect. In Proceedings of the IEEE/ACIS International Conference on Software Engineering, Artificial Intelligence, Networking and Parallel/Distributed Computing (SERA 2025), pp. 2-8, Hammamet, Tunisia, May. IEEE. [find](https://scholar.google.com/scholar?q=Baazaoui%2C+H%2E%2C+Cheniti%2DBelcadhi%2C+L%2E%2C+%26+Zrigui%2C+M%2E+%282025%29%2E+Hate+speech+detection+in+Tunisian+dialect%2E+In+Proceedings+of+the+IEEE%2FACIS+International+Conference+on+Software+Engineering%2C+Artificial+Intelli)
6. Baruah, A., Das, K., Barbhuiya, F., & Dey, K. (2020). Context-aware sarcasm detection using BERT. In Proceedings of the Second Workshop on Figurative Language Processing, pp. 83-87, Seattle, USA, July. ACL. [find](https://scholar.google.com/scholar?q=Baruah%2C+A%2E%2C+Das%2C+K%2E%2C+Barbhuiya%2C+F%2E%2C+%26+Dey%2C+K%2E+%282020%29%2E+Context%2Daware+sarcasm+detection+using+BERT%2E+In+Proceedings+of+the+Second+Workshop+on+Figurative+Language+Processing%2C+pp%2E+83%2D87%2C+Seattle%2C+USA%2C+July) [our card](https://md.research.elamur.ai/free/papers/baruah-2020-context-aware-sarcasm-detection-bert/)
7. Breiman, L. (2001). Random Machine Learning, 45(1), pp. 5-32. forests. [find](https://scholar.google.com/scholar?q=Breiman%2C+L%2E+%282001%29%2E+Random+Machine+Learning%2C+45%281%29%2C+pp%2E+5%2D32%2E+forests%2E)
8. Chen, T., & Guestrin, C. (2016). XGBoost: A scalable tree boosting system. In Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD 2016), pp. 785-794, San Francisco, USA, August. ACM. [find](https://scholar.google.com/scholar?q=Chen%2C+T%2E%2C+%26+Guestrin%2C+C%2E+%282016%29%2E+XGBoost%3A+A+scalable+tree+boosting+system%2E+In+Proceedings+of+the+22nd+ACM+SIGKDD+International+Conference+on+Knowledge+Discovery+and+Data+Mining+%28KDD+2016%29%2C+pp%2E+785%2D794)
9. Chauhan, D. S., Akhtar, M. S., Ekbal, A., & Bhattacharyya, P. (2019). Context-aware interactive attention for multimodal sentiment and emotion analysis. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP 2019), pp. 5647-5657, Hong Kong, China, November. ACL. [find](https://scholar.google.com/scholar?q=Chauhan%2C+D%2E+S%2E%2C+Akhtar%2C+M%2E+S%2E%2C+Ekbal%2C+A%2E%2C+%26+Bhattacharyya%2C+P%2E+%282019%29%2E+Context%2Daware+interactive+attention+for+multimodal+sentiment+and+emotion+analysis%2E+In+Proceedings+of+the+2019+Conference+on+Empiri)
10. Cortes, C., & Vapnik, V. (1995). Support-vector networks. Machine Learning, 20(3), pp. 273-297. [↗](https://doi.org/10.1007/bf00994018)
11. Fourati, C., Hammami, R., Latiri, C., & Haddad, H. (2024). PoliTun: Tunisian political dataset for detecting public opinions and categories orientation. In Proceedings of the 7th International Conference on Natural Language and Speech Processing (ICNLSP 2024), pp. 178-185, Antalya, Turkey, October. ACL. [find](https://scholar.google.com/scholar?q=Fourati%2C+C%2E%2C+Hammami%2C+R%2E%2C+Latiri%2C+C%2E%2C+%26+Haddad%2C+H%2E+%282024%29%2E+PoliTun%3A+Tunisian+political+dataset+for+detecting+public+opinions+and+categories+orientation%2E+In+Proceedings+of+the+7th+International+Confere)
12. Gasmi, S., Mezghani, A., & Kherallah, M. (2024). SMOTE for enhancing Tunisian hate speech detection on social media with machine learning. International Journal of Hybrid Intelligent Systems (IJHIS), 20(4), pp. 355-368. [↗](https://doi.org/10.3233/his-240012)
13. Gharbi, S., Arfaoui, H., Haddad, H., & Kchaou, M. (2021). TEET! Tunisian dataset for toxic speech detection. arXiv preprint arXiv:2110.05287. [↗](https://arxiv.org/abs/2110.05287)
14. Graves, A., Fernández, S., & Schmidhuber, J. (2005). Bidirectional LSTM networks for improved phoneme classification. In Proceedings of the International Conference on Artificial Neural Networks (ICANN 2005), pp. 799-804, Warsaw, Poland, September. Springer. [find](https://scholar.google.com/scholar?q=Graves%2C+A%2E%2C+Fern%C3%A1ndez%2C+S%2E%2C+%26+Schmidhuber%2C+J%2E+%282005%29%2E+Bidirectional+LSTM+networks+for+improved+phoneme+classification%2E+In+Proceedings+of+the+International+Conference+on+Artificial+Neural+Networks+%28ICAN)
15. Haddad, H., Mulki, H., & Oueslati, A. (2019). T-HSAB: A Tunisian hate speech and abusive dataset. In Proceedings of the International Conference on Arabic Language Processing (ICALP 2019), pp. 251-263, Hammamet, Tunisia, November. Springer. [find](https://scholar.google.com/scholar?q=Haddad%2C+H%2E%2C+Mulki%2C+H%2E%2C+%26+Oueslati%2C+A%2E+%282019%29%2E+T%2DHSAB%3A+A+Tunisian+hate+speech+and+abusive+dataset%2E+In+Proceedings+of+the+International+Conference+on+Arabic+Language+Processing+%28ICALP+2019%29%2C+pp%2E+251%2D263)
16. Kharrat, O., Mohamed, F. A., Mtimet, I., Benamor, N., & Fourati, C. (2024). HateTune: Tunisian dialect hate speech detection dataset. In Proceedings of the International Conference on Arabic Language Processing (ICALP 2023), pp. 63-73, Hammamet, Tunisia, November. Springer. [find](https://scholar.google.com/scholar?q=Kharrat%2C+O%2E%2C+Mohamed%2C+F%2E+A%2E%2C+Mtimet%2C+I%2E%2C+Benamor%2C+N%2E%2C+%26+Fourati%2C+C%2E+%282024%29%2E+HateTune%3A+Tunisian+dialect+hate+speech+detection+dataset%2E+In+Proceedings+of+the+International+Conference+on+Arabic+Language+)
17. Mekki, A., Zribi, I., Ellouze, M., & Belguith, L. H. (2022). Sarcasm detection in Tunisian social media comments: Case of COVID-19. In Proceedings of the International Symposium on Methodologies for Intelligent Systems (ISMIS 2022), pp. 45-57, Hammamet, Tunisia, October. Springer. [find](https://scholar.google.com/scholar?q=Mekki%2C+A%2E%2C+Zribi%2C+I%2E%2C+Ellouze%2C+M%2E%2C+%26+Belguith%2C+L%2E+H%2E+%282022%29%2E+Sarcasm+detection+in+Tunisian+social+media+comments%3A+Case+of+COVID%2D19%2E+In+Proceedings+of+the+International+Symposium+on+Methodologies+for+I)
18. Salomon, P. O., Kechaou, Z., & Wali, A. (2022). Arabic hate speech detection system based on AraBERT. In Proceedings of the International Conference on Computing, Intelligence and Communication (ICCICC 2022), pp. 208-213, Monastir, Tunisia, June. Springer. [find](https://scholar.google.com/scholar?q=Salomon%2C+P%2E+O%2E%2C+Kechaou%2C+Z%2E%2C+%26+Wali%2C+A%2E+%282022%29%2E+Arabic+hate+speech+detection+system+based+on+AraBERT%2E+In+Proceedings+of+the+International+Conference+on+Computing%2C+Intelligence+and+Communication+%28ICCIC)
19. Sherstinsky, A. (2018). Fundamentals of recurrent neural network (RNN) and LSTM network. arXiv preprint arXiv:1808.03314. [↗](https://arxiv.org/abs/1808.03314)
20. Shiwakoti, S., Thapa, S., Rauniyar, K., Shah, A., Bhandari, A., & Naseem, U. (2024). Analyzing the Dynamics of Climate Change Discourse on Twitter: A New Annotated Corpus and Multi-Aspect Classification. In Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024), pp. 984-994, Torino, Italy, May. ELRA and ICCL. [find](https://scholar.google.com/scholar?q=Shiwakoti%2C+S%2E%2C+Thapa%2C+S%2E%2C+Rauniyar%2C+K%2E%2C+Shah%2C+A%2E%2C+Bhandari%2C+A%2E%2C+%26+Naseem%2C+U%2E+%282024%29%2E+Analyzing+the+Dynamics+of+Climate+Change+Discourse+on+Twitter%3A+A+New+Annotated+Corpus+and+Multi%2DAspect+Classificati)
21. Tenney, I., Das, D., & Pavlick, E. (2019). BERT rediscovers the classical NLP pipeline. In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics (ACL 2019), pp. 4593-4601, Florence, Italy, July. ACL. [↗](https://doi.org/10.48550/arxiv.1905.05950) [our card](https://md.research.elamur.ai/free/papers/tenney-2019-bert-rediscovers-classical-nlp-pipeline/)
22. Trabelsi, F. B. F., & Kouki, S. (2024). Bully.tn: A cyberbullying detection dataset in Tunisian dialect. In Proceedings of the IEEE/ACS International Conference on Computer Systems and Applications (AICCSA 2024), pp. 1-8, Sousse, Tunisia, May. IEEE. [find](https://scholar.google.com/scholar?q=Trabelsi%2C+F%2E+B%2E+F%2E%2C+%26+Kouki%2C+S%2E+%282024%29%2E+Bully%2Etn%3A+A+cyberbullying+detection+dataset+in+Tunisian+dialect%2E+In+Proceedings+of+the+IEEE%2FACS+International+Conference+on+Computer+Systems+and+Applications+%28A)