---
title: "ArabicDialectSafety — Zaghouani 2026"
description: "Arabic safety filters are usually built and tested on Modern Standard Arabic."
authors: ["Wajdi Zaghouani", "Md. Rafiul Biswas", "Kholoud Khalil Aldous", "Mabrouka Bessghaier"]
year: 2026
venue: "arXiv:2608.01291 · EMNLP 2026 Findings"
source: "https://arxiv.org/abs/2608.01291"
doi: "10.48550/arxiv.2608.01291"
arxiv: "2608.01291"
licence: "CC BY 4.0 (arXiv)"
licence_url: "http://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/zaghouani-2026-arabicdialectsafety-dialect-aware-benchmark-arabic/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/zaghouani-2026-arabicdialectsafety-dialect-aware-benchmark-arabic/)

[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/) / ArabicDialectSafety: A Dialect-Aware Benchmark for Arabic Content Safety Classification

# ArabicDialectSafety: A Dialect-Aware Benchmark for Arabic Content Safety Classification

**Authors:** [Wajdi Zaghouani](https://md.research.elamur.ai/free/en/authors/wajdi-zaghouani/), [Md. Rafiul Biswas](https://md.research.elamur.ai/free/en/authors/md-rafiul-biswas/), [Kholoud Khalil Aldous](https://md.research.elamur.ai/free/en/authors/kholoud-khalil-aldous/), [Mabrouka Bessghaier](https://md.research.elamur.ai/free/en/authors/mabrouka-bessghaier/) · **Year:** 2026 · **Venue:** arXiv:2608.01291 · EMNLP 2026 Findings · **Licence:** [CC BY 4.0 (arXiv)](http://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://arxiv.org/abs/2608.01291) · [PDF](https://arxiv.org/pdf/2608.01291) · [DOI 10.48550/arxiv.2608.01291](https://doi.org/10.48550/arxiv.2608.01291) · [arXiv 2608.01291](https://arxiv.org/abs/2608.01291) · [OpenAlex](https://openalex.org/W7172431989)

**Citations:** 0 citations (openalex-snapshot · 2026-10-14) · [1](#cited-by) in our library, of the 140884 papers whose bibliographies we read

## Summary
Arabic safety filters are usually built and tested on Modern Standard Arabic. Harmful requests phrased in regional dialects can therefore slip past them. Earlier Arabic safety resources either ignore dialects, mix them without labels, or are too small to evaluate each dialect separately.

The authors built ArabicDialectSafety, a set of 24,051 prompts written by native-speaker annotators. It covers MSA plus Syrian, Egyptian, Algerian, Palestinian and Moroccan, and each prompt is labeled with its dialect, one of seven harm categories, and a safe/unsafe tag. They define two tasks, binary detection and fine-grained harm classification, and report results per dialect.

Among seven classifiers, a fine-tuned MARBERTv2 performs best and clearly beats prompted LLMs, including Arabic-specialized ones. Telling a model which dialect it is reading helps only when that information enters the model's internal representation, not when it is stated in the prompt. Maghrebi varieties remain noticeably harder. When frontier LLMs are given harmful dialectal prompts, they rarely produce unsafe answers, though the automatic judge used to measure this seems to undercount.

## Key points
- The dataset has 24,051 human-authored prompts across six varieties and seven harm categories. Only Self-Harming, Harming Others and Harassment include benign counterpart prompts; the other four categories (Adult Content, Hate Speech, Bullying, and Fraud & Deception) consist entirely of unsafe examples.
- Inter-annotator agreement before adjudication (Cohen's κ) was 0.91 for dialect, 0.88 for safe/unsafe and 0.81 for harm category. Harassment was lowest at 0.72, and about 11% of items needed discussion or adjudication.
- The evaluated systems were TF-IDF + LR, multilingual-e5-base, MARBERTv2, FANAR-9B-Instruct, Qwen3-8B, Llama-3.1-8B and Jais-2-8B. MARBERTv2 reached 0.95 binary and 0.90 granular Macro-F1.
- Frontier LLMs produced unsafe answers in under 5% of cases, as scored by a GPT-5.2 judge. A human check of 242 prompt–response pairs agreed with the judge 77.7% of the time (κ = 0.44), and the judge tended to under-flag unsafe outputs.
- Compared with AraSafe, the dataset is about twice as large and uses manual, country-level dialect labels rather than automatic regional groupings. Access is controlled and limited to non-commercial research.

## Interesting ideas
- Annotators wrote each harmful prompt alongside a benign counterpart on the same topic, so a classifier cannot succeed just by spotting the topic. — For Mova (learn): Pairing each test item with a near-identical control whose correct answer differs would make Mova's evaluations show whether a component reasons or only matches surface cues.
- Agreement was reported separately for each category, which showed exactly where the taxonomy overlaps, such as bullying vs. harming others vs. harassment. — For Mova (frames): Measuring agreement per label for speech-act and semantic-role tags would show which frame distinctions are genuinely unclear and should be merged or more sharply defined.
- Dialect information helped only when built into the model's representation; writing it into the prompt did not help. — For Mova (tree): gparse could receive the treebank variety or genre as a learned input embedding rather than a textual hint, which may help when training on converted UD dialects.
- The authors checked the LLM judge against a human-labeled sample, found that its errors lean in one direction, and therefore reported its rates as a lower bound. — For Mova (learn): Each LLM teacher tier could be calibrated against a small human-checked slice, recording not just accuracy but which way it errs before its silver labels are trusted.
- Splitting every result by variety revealed large gaps for low-resource dialects that a single overall score would hide. — For Mova (tree): Reporting LAS separately for each source treebank or genre, rather than one pooled number, would expose which domains the parsers quietly fail on.

*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/en/shelves/linguistics/)

Topics: [Evaluation & benchmarks](https://md.research.elamur.ai/free/en/topics/evaluation-benchmarks/) · [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · [Bias, fairness & safety](https://md.research.elamur.ai/free/en/topics/bias-fairness-safety/)

## Cited by

Cited by

1 paper of our library cites it (of the 140884 whose bibliographies we read); each has a page here.

- [ArGuard Shared Task: Harmful Content Detection in Arabic Memes and LLM Prompts](https://md.research.elamur.ai/free/en/papers/alam-2026-arguard-shared-task-harmful-content/)

## Cite

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

```bibtex
@misc{zaghouani2026arabicdialectsafety,
  title = {{ArabicDialectSafety: A Dialect-Aware Benchmark for Arabic Content Safety Classification}},
  author = {Zaghouani, Wajdi and Biswas, Md. Rafiul and Aldous, Kholoud Khalil and Bessghaier, Mabrouka},
  year = {2026},
  doi = {10.48550/arxiv.2608.01291},
  eprint = {2608.01291},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2608.01291}
}
```

- [← previous Arabic Sentence Segmentation Across Genres and Punctuation Conditions](https://md.research.elamur.ai/free/en/papers/langstream-2026-elkholy-arabic-sentence-segmentation-across-genres/)
- [next → Are LLMs Good Text Diacritizers? An Arabic and Yoruba Case Study](https://md.research.elamur.ai/free/en/papers/lingbulk-2026-toyin-are-llms-good-text-diacritizers/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/)

## References

32 references, 14 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Ahmed Abdelali, Hamdy Mubarak, Younes Samih, Sabit Hassan, and Kareem Darwish. 2021. QADI: Arabic dialect identification in the wild. In Proceedings of the Sixth Arabic Natural Language Processing Workshop, pages 1–10, Kyiv, Ukraine (Virtual). Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Ahmed+Abdelali%2C+Hamdy+Mubarak%2C+Younes+Samih%2C+Sabit+Hassan%2C+and+Kareem+Darwish%2E+2021%2E+QADI%3A+Arabic+dialect+identification+in+the+wild%2E+In+Proceedings+of+the+Sixth+Arabic+Natural+Language+Processing+Wor) [our card](https://md.research.elamur.ai/free/en/papers/abdelali-2020-arabic-dialect-identification-wild/)
2. Muhammad Abdul-Mageed, AbdelRahim Elmadany, and El Moatez Billah Nagoudi. 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 (Volume 1: Long Papers), pages 7088–7105, Online. Association for Computational Linguistics. [↗](https://doi.org/10.18653/v1/2021.acl-long.551) [our card](https://md.research.elamur.ai/free/en/papers/abdul-2021-mageed-arbert-marbert-deep-bidirectional-transformers/)
3. Emad A. Alghamdi, Reem I. Masoud, Deema Alnuhait, Afnan Y. Alomairi, Ahmed Ashraf, and Mohamed Zaytoon. 2025. AraTrust: An evaluation of trustworthiness for LLMs in Arabic. In Proceedings of the 31st International Conference on Computational Linguistics, pages 8664– 8679, Abu Dhabi, UAE. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Emad+A%2E+Alghamdi%2C+Reem+I%2E+Masoud%2C+Deema+Alnuhait%2C+Afnan+Y%2E+Alomairi%2C+Ahmed+Ashraf%2C+and+Mohamed+Zaytoon%2E+2025%2E+AraTrust%3A+An+evaluation+of+trustworthiness+for+LLMs+in+Arabic%2E+In+Proceedings+of+the+31st+)
4. Yasser Ashraf, Yuxia Wang, Bin Gu, Preslav Nakov, and Timothy Baldwin. 2025. Arabic dataset for LLM safeguard evaluation. In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), pages 5529–5546, Albuquerque, New Mexico. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Yasser+Ashraf%2C+Yuxia+Wang%2C+Bin+Gu%2C+Preslav+Nakov%2C+and+Timothy+Baldwin%2E+2025%2E+Arabic+dataset+for+LLM+safeguard+evaluation%2E+In+Proceedings+of+the+2025+Conference+of+the+Nations+of+the+Americas+Chapter+o)
5. Somnath Banerjee, Pratyush Chatterjee, Shanu Kumar, Sayan Layek, Parag Agrawal, Rima Hazra, and Animesh Mukherjee. 2025. Attributional safety failures in large language models under code-mixed perturbations. Preprint, arXiv:2505.14469. [↗](https://arxiv.org/abs/2505.14469)
6. Houda Bouamor, Nizar Habash, Mohammad Salameh, Wajdi Zaghouani, Owen Rambow, Dana Abdulrahim, Ossama Obeid, Salam Khalifa, Fadhl Eryani, Alexander Erdmann, and The MADAR Ara-Kemal Oflazer. 2018. bic dialect corpus and lexicon. In Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), Miyazaki, Japan. European Language Resources Association (ELRA). [find](https://scholar.google.com/scholar?q=Houda+Bouamor%2C+Nizar+Habash%2C+Mohammad+Salameh%2C+Wajdi+Zaghouani%2C+Owen+Rambow%2C+Dana+Abdulrahim%2C+Ossama+Obeid%2C+Salam+Khalifa%2C+Fadhl+Eryani%2C+Alexander+Erdmann%2C+and+The+MADAR+Ara%2DKemal+Oflazer%2E+2018%2E+bic+d)
7. Adrian de Wynter, Ishaan Watts, Tua Wongsangaroonsri, Minghui Zhang, Noura Farra, Nektar Ege Altıntoprak, Lena Baur, Samantha Claudet, Pavel Gajdušek, Qilong Gu, Anna Kaminska, Tomasz Kaminski, Ruby Kuo, Akiko Kyuba, Jongho Lee, Kartik Mathur, Petter Merok, Ivana Milovanović, Nani Paananen, and 13 others. 2025. RTP-LX: Can LLMs evaluate toxicity in multilingual scenarios? In Proceedings of the Thirty-Ninth AAAI Conference on Artificial Intelligence and Thirty-Seventh Conference on Innovative Applications of Artificial Intelligence and Fifteenth Symposium on Educational Advances in Artificial Intelligence, AAAI’25/IAAI’25/EAAI’25. AAAI Press. [find](https://scholar.google.com/scholar?q=Adrian+de+Wynter%2C+Ishaan+Watts%2C+Tua+Wongsangaroonsri%2C+Minghui+Zhang%2C+Noura+Farra%2C+Nektar+Ege+Alt%C4%B1ntoprak%2C+Lena+Baur%2C+Samantha+Claudet%2C+Pavel+Gajdu%C5%A1ek%2C+Qilong+Gu%2C+Anna+Kaminska%2C+Tomasz+Kaminski%2C+Ruby+K)
8. Masoomali Fatehkia, Enes Altinisik, and Husrev Taha Sencar. 2026. FanarGuard: A culturally-aware moderation filter for Arabic language models. pages 7848–7869. [find](https://scholar.google.com/scholar?q=Masoomali+Fatehkia%2C+Enes+Altinisik%2C+and+Husrev+Taha+Sencar%2E+2026%2E+FanarGuard%3A+A+culturally%2Daware+moderation+filter+for+Arabic+language+models%2E+pages+7848%E2%80%937869%2E)
9. Samuel Gehman, Suchin Gururangan, Maarten Sap, Yejin Choi, and Noah A. Smith. 2020. RealToxicityPrompts: Evaluating neural toxic degeneration in language models. In Findings of the Association for Computational Linguistics: EMNLP 2020, pages 3356–3369, Online. Association for Computational Linguistics. [↗](https://doi.org/10.18653/v1/2020.findings-emnlp.301) [our card](https://md.research.elamur.ai/free/en/papers/gehman-2020-realtoxicityprompts-evaluating-neural-toxic-degeneration/)
10. Thomas Hartvigsen, Saadia Gabriel, Hamid Palangi, Maarten Sap, Dipankar Ray, and Ece Kamar. 2022. ToxiGen: A large-scale machine-generated dataset for adversarial and implicit hate speech detection. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 3309–3326, Dublin, Ireland. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Thomas+Hartvigsen%2C+Saadia+Gabriel%2C+Hamid+Palangi%2C+Maarten+Sap%2C+Dipankar+Ray%2C+and+Ece+Kamar%2E+2022%2E+ToxiGen%3A+A+large%2Dscale+machine%2Dgenerated+dataset+for+adversarial+and+implicit+hate+speech+detection%2E+I) [our card](https://md.research.elamur.ai/free/en/papers/hartvigsen-2022-toxigen-large-scale-machine-generated/)
11. Sabit Hassan, Hamdy Mubarak, Ahmed Abdelali, and Kareem Darwish. 2021. ASAD: Arabic social media analytics and unDerstanding. In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: System Demonstrations, pages 113–118, Online. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Sabit+Hassan%2C+Hamdy+Mubarak%2C+Ahmed+Abdelali%2C+and+Kareem+Darwish%2E+2021%2E+ASAD%3A+Arabic+social+media+analytics+and+unDerstanding%2E+In+Proceedings+of+the+16th+Conference+of+the+European+Chapter+of+the+Assoc)
12. Hakan Inan, Kartikeya Upasani, Jianfeng Chi, Rashi Rungta, Krithika Iyer, Yuning Mao, Michael Tontchev, Qing Hu, Brian Fuller, Davide Testuggine, and Madian Khabsa. 2023. Llama guard: Llm-based input-output safeguard for human-ai conversations. Preprint, arXiv:2312.06674. [↗](https://arxiv.org/abs/2312.06674) [our card](https://md.research.elamur.ai/free/en/papers/inan-2023-llama-guard-llm-based-input/)
13. Pratik Jalan, Vadivel Abishethvarman, Bhavik Chandna, and Usman Naseem. 2026. Survey on llm safety: Attacks, defenses, alignment, metrics, and guardrails. Machine Learning, 115(6):130. [find](https://scholar.google.com/scholar?q=Pratik+Jalan%2C+Vadivel+Abishethvarman%2C+Bhavik+Chandna%2C+and+Usman+Naseem%2E+2026%2E+Survey+on+llm+safety%3A+Attacks%2C+defenses%2C+alignment%2C+metrics%2C+and+guardrails%2E+Machine+Learning%2C+115%286%29%3A130%2E)
14. Priyanshu Kumar, Devansh Jain, Akhila Yerukola, Liwei Jiang, Himanshu Beniwal, Thomas Hartvigsen, and Maarten Sap. 2025. Polyguard: A multilingual safety moderation tool for 17 languages. Preprint, arXiv:2504.04377. [↗](https://arxiv.org/abs/2504.04377)
15. JR Landis and GG Koch. 1977. The measurement of observer agreement for categorical data. Biometrics, 33(1):159—174. [find](https://scholar.google.com/scholar?q=JR+Landis+and+GG+Koch%2E+1977%2E+The+measurement+of+observer+agreement+for+categorical+data%2E+Biometrics%2C+33%281%29%3A159%E2%80%94174%2E)
16. Zi Lin, Zihan Wang, Yongqi Tong, Yangkun Wang, Yuxin Guo, Yujia Wang, and Jingbo Shang. 2023. ToxicChat: Unveiling hidden challenges of toxicity detection in real-world user-AI conversation. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 4694–4702, Singapore. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Zi+Lin%2C+Zihan+Wang%2C+Yongqi+Tong%2C+Yangkun+Wang%2C+Yuxin+Guo%2C+Yujia+Wang%2C+and+Jingbo+Shang%2E+2023%2E+ToxicChat%3A+Unveiling+hidden+challenges+of+toxicity+detection+in+real%2Dworld+user%2DAI+conversation%2E+In+Findin) [our card](https://md.research.elamur.ai/free/en/papers/lin-2023-toxicchat-unveiling-hidden-challenges-toxicity/)
17. Malak Mashabi, Shahad Al-Khalifa, and Hend Al-Khalifa. 2026. A survey of large language models for arabic language and its dialects. ACM Trans. Asian Low-Resour. Lang. Inf. Process., 25(7). [find](https://scholar.google.com/scholar?q=Malak+Mashabi%2C+Shahad+Al%2DKhalifa%2C+and+Hend+Al%2DKhalifa%2E+2026%2E+A+survey+of+large+language+models+for+arabic+language+and+its+dialects%2E+ACM+Trans%2E+Asian+Low%2DResour%2E+Lang%2E+Inf%2E+Process%2E%2C+25%287%29%2E) [our card](https://md.research.elamur.ai/free/en/papers/zhao-2023-survey-large-language-models/)
18. Hamdy Mubarak, Hend Al-Khalifa, and Abdulmohsen Al-Thubaity. 2022. Overview of OS-ACT5 shared task on Arabic offensive language and hate speech detection. In Proceedinsg of the 5th Workshop on Open-Source Arabic Corpora and Processing Tools with Shared Tasks on Qur’an QA and Fine-Grained Hate Speech Detection, pages 162–166, Marseille, France. European Language Resources Association. [find](https://scholar.google.com/scholar?q=Hamdy+Mubarak%2C+Hend+Al%2DKhalifa%2C+and+Abdulmohsen+Al%2DThubaity%2E+2022%2E+Overview+of+OS%2DACT5+shared+task+on+Arabic+offensive+language+and+hate+speech+detection%2E+In+Proceedinsg+of+the+5th+Workshop+on+Open%2DSo)
19. Hamdy Mubarak, Kareem Darwish, and Walid Magdy. 2017. Abusive language detection on Arabic social media. In Proceedings of the First Workshop on Abusive Language Online, pages 52–56, Vancouver, BC, Canada. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Hamdy+Mubarak%2C+Kareem+Darwish%2C+and+Walid+Magdy%2E+2017%2E+Abusive+language+detection+on+Arabic+social+media%2E+In+Proceedings+of+the+First+Workshop+on+Abusive+Language+Online%2C+pages+52%E2%80%9356%2C+Vancouver%2C+BC%2C+Ca) [our card](https://md.research.elamur.ai/free/en/papers/mubarak-2017-abusive-language-detection-arabic-social/)
20. Hamdy Mubarak, Kareem Darwish, Walid Magdy, Tamer Elsayed, and Hend Al-Khalifa. 2020. Overview of OSACT4 Arabic offensive language detection shared task. In Proceedings of the 4th Workshop on Open-Source Arabic Corpora and Processing Tools, with a Shared Task on Offensive Language Detection, pages 48–52, Marseille, France. European Language Resources Association. [find](https://scholar.google.com/scholar?q=Hamdy+Mubarak%2C+Kareem+Darwish%2C+Walid+Magdy%2C+Tamer+Elsayed%2C+and+Hend+Al%2DKhalifa%2E+2020%2E+Overview+of+OSACT4+Arabic+offensive+language+detection+shared+task%2E+In+Proceedings+of+the+4th+Workshop+on+Open%2DSou)
21. Hamdy Mubarak, Abubakr Mohamed, and Majd Hawasly. 2025. AraSafe: Benchmarking safety in Arabic LLMs. In Findings of the Association for Computational Linguistics: EMNLP 2025, pages 9976–9992, Suzhou, China. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Hamdy+Mubarak%2C+Abubakr+Mohamed%2C+and+Majd+Hawasly%2E+2025%2E+AraSafe%3A+Benchmarking+safety+in+Arabic+LLMs%2E+In+Findings+of+the+Association+for+Computational+Linguistics%3A+EMNLP+2025%2C+pages+9976%E2%80%939992%2C+Suzhou%2C+)
22. Hala Mulki, Hatem Haddad, Chedi Bechikh Ali, and Halima Alshabani. 2019. L-HSAB: A Levantine Twitter dataset for hate speech and abusive language. In Proceedings of the Third Workshop on Abusive Language Online, pages 111– 118, Florence, Italy. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Hala+Mulki%2C+Hatem+Haddad%2C+Chedi+Bechikh+Ali%2C+and+Halima+Alshabani%2E+2019%2E+L%2DHSAB%3A+A+Levantine+Twitter+dataset+for+hate+speech+and+abusive+language%2E+In+Proceedings+of+the+Third+Workshop+on+Abusive+Langu)
23. Ossama Obeid, Nasser Zalmout, Salam Khalifa, Dima Taji, Mai Oudah, Bashar Alhafni, Go Inoue, Fadhl Eryani, Alexander Erdmann, and Nizar Habash. 2020. CAMeL tools: An open source python toolkit for Arabic natural language processing. In Proceedings of the Twelfth Language Resources and Evaluation Conference, pages 7022–7032, Marseille, France. European Language Resources Association. [find](https://scholar.google.com/scholar?q=Ossama+Obeid%2C+Nasser+Zalmout%2C+Salam+Khalifa%2C+Dima+Taji%2C+Mai+Oudah%2C+Bashar+Alhafni%2C+Go+Inoue%2C+Fadhl+Eryani%2C+Alexander+Erdmann%2C+and+Nizar+Habash%2E+2020%2E+CAMeL+tools%3A+An+open+source+python+toolkit+for+Ara) [our card](https://md.research.elamur.ai/free/en/papers/obeid-2020-camel-tools-open-source-python/)
24. Soufiayn Ouali, Kanza Raisi, Asmaa Mourhir, El Habib Nfaoui, and Said EI Garouani. 2026. Adversarial evaluation of large language models for building robust offensive language detection in moroccan arabic. Big Data and Cognitive Computing. Neha Sengupta, Sunil Kumar Sahu, Bokang Jia, Satheesh Katipomu, Haonan Li, Fajri Koto, William Marshall, Gurpreet Gosal, Cynthia Liu, Zhiming Chen, Osama Mohammed Afzal, Samta Kamboj, Onkar Pandit, Rahul Pal, Lalit Pradhan, Zain Muhammad Mujahid, Massa Baali, Xudong Han, Sondos Mahmoud Bsharat, and 13 others. 2023. Jais and jais-chat: Arabic-centric foundation and instruction-tuned open generative large language models. Preprint, arXiv:2308.16149. [↗](https://arxiv.org/abs/2308.16149) [our card](https://md.research.elamur.ai/free/en/papers/sengupta-2023-jais-jais-chat-arabic-centric/)
25. Fanar Team, Ummar Abbas, Mohammad Shahmeer Ahmad, Firoj Alam, Enes Altinisik, Ehsannedin Asgari, Yazan Boshmaf, Sabri Boughorbel, Sanjay Chawla, Shammur Chowdhury, Fahim Dalvi, Kareem Darwish, Nadir Durrani, Mohamed Elfeky, Ahmed Elmagarmid, Mohamed Eltabakh, Masoomali Fatehkia, Anastasios Fragkopoulos, Maram Hasanain, and 23 others. 2025. Fanar: An arabic-centric multimodal generative ai platform. Preprint, arXiv:2501.13944. [↗](https://arxiv.org/abs/2501.13944) [our card](https://md.research.elamur.ai/free/en/papers/abbas-2025-fanar-arabic-centric-multimodal-generative/)
26. Hugo Touvron, Thibaut Lavril, Gautier Izacard, Xavier Martinet, Marie-Anne Lachaux, Timothée Lacroix, Baptiste Rozière, Naman Goyal, Eric Hambro, Faisal Azhar, Aurelien Rodriguez, Armand Joulin, Edouard Grave, and Guillaume Lample. 2023. Llama: Open and efficient foundation language models. Preprint, arXiv:2302.13971. [↗](https://arxiv.org/abs/2302.13971)
27. Liang Wang, Nan Yang, Xiaolong Huang, Linjun Yang, Rangan Majumder, and Furu Wei. 2024a. Multilingual e5 text embeddings: A technical report. Preprint, arXiv:2402.05672. [↗](https://arxiv.org/abs/2402.05672) [our card](https://md.research.elamur.ai/free/en/papers/wang-2024-multilingual-e5-text-embeddings-technical/)
28. Yuxia Wang, Zenan Zhai, Haonan Li, Xudong Han, Shom Lin, Zhenxuan Zhang, Angela Zhao, Preslav Nakov, and Timothy Baldwin. 2024b. A Chinese dataset for evaluating the safeguards in large language models. In Findings of the Association for Computational Linguistics: ACL 2024, pages 3106–3119, Bangkok, Thailand. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Yuxia+Wang%2C+Zenan+Zhai%2C+Haonan+Li%2C+Xudong+Han%2C+Shom+Lin%2C+Zhenxuan+Zhang%2C+Angela+Zhao%2C+Preslav+Nakov%2C+and+Timothy+Baldwin%2E+2024b%2E+A+Chinese+dataset+for+evaluating+the+safeguards+in+large+language+model)
29. An Yang, Anfeng Li, Baosong Yang, Beichen Zhang, Binyuan Hui, Bo Zheng, Bowen Yu, Chang Gao, Chengen Huang, Chenxu Lv, Chujie Zheng, Dayiheng Liu, Fan Zhou, Fei Huang, Feng Hu, Hao Ge, Haoran Wei, Huan Lin, Jialong Tang, and 41 others. 2025a. Qwen3 technical report. Preprint, arXiv:2505.09388. [↗](https://arxiv.org/abs/2505.09388) [our card](https://md.research.elamur.ai/free/en/papers/yang-2025-qwen3-technical-report/)
30. Yahan Yang, Soham Dan, Shuo Li, Dan Roth, and Insup Lee. 2025b. MrGuard: A multilingual reasoning guardrail for universal LLM safety. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 27377–27396, Suzhou, China. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Yahan+Yang%2C+Soham+Dan%2C+Shuo+Li%2C+Dan+Roth%2C+and+Insup+Lee%2E+2025b%2E+MrGuard%3A+A+multilingual+reasoning+guardrail+for+universal+LLM+safety%2E+In+Proceedings+of+the+2025+Conference+on+Empirical+Methods+in+Natu)
31. Hamza Youseef, Luis Gonzaga Baca Ruiz, David Criado Ramón, and María del Carmen Pegala-jar Jimenez. 2026. A short review of arabic aspect-based sentiment analysis: Methods, challenges and future directions. AI, 7(4). [find](https://scholar.google.com/scholar?q=Hamza+Youseef%2C+Luis+Gonzaga+Baca+Ruiz%2C+David+Criado+Ram%C3%B3n%2C+and+Mar%C3%ADa+del+Carmen+Pegala%2Djar+Jimenez%2E+2026%2E+A+short+review+of+arabic+aspect%2Dbased+sentiment+analysis%3A+Methods%2C+challenges+and+future+direc)
32. Zhexin Zhang, Leqi Lei, Lindong Wu, Rui Sun, Yongkang Huang, Chong Long, Xiao Liu, Xuanyu Lei, Jie Tang, and Minlie Huang. 2024. SafetyBench: Evaluating the safety of large language models. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 15537–15553, Bangkok, Thailand. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Zhexin+Zhang%2C+Leqi+Lei%2C+Lindong+Wu%2C+Rui+Sun%2C+Yongkang+Huang%2C+Chong+Long%2C+Xiao+Liu%2C+Xuanyu+Lei%2C+Jie+Tang%2C+and+Minlie+Huang%2E+2024%2E+SafetyBench%3A+Evaluating+the+safety+of+large+language+models%2E+In+Proceed) [our card](https://md.research.elamur.ai/free/en/papers/zhang-2023-safetybench-evaluating-safety-large/)