Elamur Research⌕
☰
home / papers / 2026 / Why Current XAI Is Not Enough for Arabic NLP: A Critical Survey of the Explainability Gap

Why Current XAI Is Not Enough for Arabic NLP: A Critical Survey of the Explainability Gap

Paper · 2026 · Salima Lamsiyah and Ruslan Mitkov

Summary

This critical survey asks why explainability work on Arabic NLP still falls short, even as Arabic transformers and LLMs keep improving. The authors searched ACL Anthology, IEEE Xplore, arXiv, Google Scholar and other indexes. They kept only Arabic-related studies with an explicit explanatory component and coded each one for its task, model family and XAI method, the Arabic variety or register it dealt with, what its explanations were meant to achieve, how those explanations were evaluated, and how deeply they drew on linguistics. From this they identify three linked shortfalls:

  • Method gap: a narrow toolkit of mostly LIME, SHAP, attention and saliency.
  • Task gap: a heavy focus on classification such as sentiment, hate speech, fake news and spam.
  • Linguistic gap: highlighted tokens rarely say anything about Arabic morphology, clitics, dialects, diacritics or religious registers.

They also propose four levels an explanation may target (prediction, model, linguistic, socio-cultural), a taxonomy with a task-method coverage table, and a research agenda. The main claim is that explanations must be faithful to Arabic as a language and a cultural object, not just to a model's output. Few papers qualified for the survey, which itself shows how underdeveloped this area is.

Key points

  • The four explanation levels are prediction-level, model-level, linguistic and socio-cultural. Most Arabic XAI papers stay at the prediction level, using local feature attribution.
  • The coverage table shows that sentiment and harmful-content classification dominate. Generation, RAG, translation, summarization, parsing, dialogue and structured analysis are barely covered, and method coverage is even narrower than task coverage.
  • A few studies show richer directions:
    • Abdelali et al. (2022) probe Arabic transformers for morphology, syntax and dialect.
    • Sahyoun and Shehata (2023) design explainable metrics for dialectal ASR.
    • Younes (2026) builds visual analytics for Arabic NER.
    • AlGhamdi et al. (2026) use knowledge-graph diagnostics for hallucinations in Qur'anic MRC.
  • The survey treats attention-based models with caution, because attention weights are not necessarily an explanation.
  • The agenda includes:
    • Arabic-aware explanation units
    • cross-variety evaluation and perturbation tests
    • human rationale benchmarks
    • LLM/RAG explanation protocols
    • reporting of normalization, segmentation, diacritic handling, checkpoints and explainer parameters

Interesting ideas

  • I1. An explanation should point at the real linguistic unit (stem, clitic, lemma, multiword expression, orthographic variant), not at a whitespace token; otherwise a morphological cause looks like a word-level one.
  • I2. Label each explanation with its level (output, internal mechanism, linguistic, socio-cultural), so it is clear what kind of claim is being made and how to test it.
  • I3. Report faithfulness, plausibility to humans, stability under meaning-preserving perturbations, usefulness and fairness as separate measurements. A convincing explanation is not automatically a faithful one.
  • I4. Diagnose hallucinations by aligning a generated answer with structured triples from a knowledge graph, instead of judging the answer text alone.
  • I5. Without full reporting of preprocessing, segmentation and normalization, nobody can tell whether a method is genuinely better or just easier to visualize.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (tokens): Attach tagger and lemmatizer decisions to morpheme and lemma evidence, so later explanations can say which affix or lemma triggered a tag.
  • I2 (answer): Each answer explanation could state whether it cites the frame, the world-state fact or background knowledge, making the reasoning chain auditable.
  • I3 (learn): Add perturbation suites (paraphrase, spelling variants, dialect-like rewrites) as negative controls, and check that explanations stay stable when meaning stays fixed.
  • I4 (world): Check every answer against the triples in the world of the text, and flag any claim with no supporting entity, state or event.
  • I5 (tree): Log the tokenization and UD dialect-conversion settings with every parser evaluation, so that LAS differences can be attributed to the parser rather than to the preprocessing.

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

🎓University of AlicanteAlicante · Spain21 works · 1,098 citations🎓University of LuxembourgLuxembourg · Luxembourg27 works · 337 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 2Research subjects and languages the work is about.Semitic languages · 2,908 worksInterpretability · 130 works

Cite

Cite as Salima Lamsiyah and Ruslan Mitkov. 2026. Why Current XAI Is Not Enough for Arabic NLP: A Critical Survey of the Explainability Gap. arXiv:2608.26144.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.48550/arxiv.2608.26144 · arXiv 2608.26144

@misc{lamsiyah2026why,
  title = {{Why Current XAI Is Not Enough for Arabic NLP: A Critical Survey of the Explainability Gap}},
  author = {Salima Lamsiyah and Ruslan Mitkov},
  year = {2026},
  doi = {10.48550/arxiv.2608.26144},
  eprint = {2608.26144},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.48550/arxiv.2608.26144}
}

References

43 references, 13 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Ahmed Abdelali, Nadir Durrani, Fahim Dalvi, and Hassan Sajjad. 2022. Post-hoc analysis of arabic transformer models. In Proceedings of the Fifth BlackboxNLP Workshop on Analyzing and Interpreting Neural Networks for NLP, pages 91–103.
  2. Youmna Abdelwahab, Mohamed Kholief, and Ahmed Ahmed Hesham Sedky. 2022. Justifying arabic text sentiment analysis using explainable ai (xai): Lasik surgeries case study. Information, 13(11):536. ↗
  3. Muhammad Abdul-Mageed, Abdel Rahim 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. ↗
  4. Noor Al-Ansari, Dena Al-Thani, and Mariam Ba-hameish. 2026. Explaining in context: Perceived informativeness of explainable artificial intelligence (xai) in arabic hate speech detection. Computers in Human Behavior Reports, page 101083. ↗
  5. Aisha Alansari, Dorieh Alomari, Sajjad Mahmood, and Irfan Ahmad. 2026. Multi-label classification of arabic app reviews with data augmentation and explainable ai. Arabian Journal for Science and Engineering, 51(9):12487–12506. ↗
  6. Najwa Abdullah AlGhamdi, Sadam Al-Azani, Kwabena Nuamah, and Alan Bundy. 2026. A knowledge graph based diagnostic framework for analyzing hallucinations in arabic machine reading comprehension. In Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script, pages 413–421.
  7. Turki Aljrees. 2024. Improving prediction of arabic fake news using elmo’s features-based tri-ensemble model and lime xai. IEEE Access, 12:63066–63076. ↗
  8. Afnan Alsehaimi, Amal Babour, and Dimah Alahmadi. 2025. Toward transparent modeling: A scoping review of explainability for arabic sentiment analysis. Applied Sciences, 15(19):10659. ↗
  9. Bunder Sebail Alshammari, Khaled Nasser Alfraidi, Mohammad Omar Aljudaiey, and Walid Abdelhalim. 2025. Robustness and explainability in arabic authorship models: Tackling the challenge of ai-generated texts. ↗
  10. Majed Alwateer, Ibrahim Gad, Mahmoud Elmarhomy, Ghada Elmarhomy, Hasan Hashim, Malik Almaliki, and El-Sayed Atlam. 2025. Interpretable arabic hate speech detection using large language model. In 2025 2nd International Conference on Advanced Innovations in Smart Cities, pages 1–8. IEEE. find
  11. Md Atabuzzaman, Md Shajalal, Maksuda Bilkis Baby, and Alexander Boden. 2023. Arabic sentiment analysis with noisy deep explainable model. In Proceedings of the 2023 7th International Conference on Natural Language Processing and Information Retrieval, pages 185–189. ↗
  12. Mohamed Salah Awadallah, Francisco de Arriba-Perez, Enrique Costa-Montenegro, Mohamed Kholief, and Nashwa El-Bendary. 2022. Investigation of local interpretable model-agnostic explanations (lime) framework with multi-dialect arabic text sentiment classification. In 2022 32nd International Conference on Computer Theory and Applications, pages 116–121. IEEE. find
  13. Yousra Chahinez Hadj Azzem, Fouzi Harrag, and Ladjel Bellatreche. 2024. Exploring explainability in arabic language models: An empirical analysis of techniques. Procedia Computer Science, 244:212– 219. find
  14. Nadiah A. Baghdadi, Yousry M. Abdulazeem, Hanaa ZainEldin, Tamer Ahmed Farrag, Mansourah K. Aljohani, Amer Malki, Mahmoud Badawy, and Mostafa A. Elhosseini. 2024. Toward robust arabic sign language recognition via vision transformers and local interpretable model-agnostic explanations integration. Journal of Disability Research, 3(8):20240092. ↗
  15. Mazen Balat, Rewaa Awaad, Ahmed B. Zaky, and Salah A. Aly. 2025. Revolutionizing communication with deep learning and xai for enhanced arabic sign language recognition. arXiv preprint arXiv:2501.08169. ↗
  16. Mohamed Berrimi, Mourad Oussalah, Abdelouahab Moussaoui, and Mohamed Saidi. 2023. Attention mechanism architecture for arabic sentiment analysis. ACM Transactions on Asian and Low-Resource Language Information Processing, 22(4):1–26. ↗
  17. Mohamed Aly Bouke, Omar Imhemed Alramli, Alsadg Ahmed Albadwi Abdalhafid, and Azizol Abdullah. 2026. A novel lightgbm model for arabic spam detection integrated with xai for enhanced explainability. Computers and Electrical Engineering, 133:111032. ↗
  18. Youssef Chafiqui, Houda Anoun, Latifa Aloiradi, and Fatima Ezzahraa Nazih. 2024. Xs2a: A platform for explainable sentiment analysis in arabic comments. In International Conference on Advances in Communication Technology and Computer Engineering, pages 309–320. Springer. find
  19. Kareem Darwish, Nizar Habash, Mourad Abbas, Hend Al-Khalifa, Huseein T Al-Natsheh, Houda Bouamor, Karim Bouzoubaa, Violetta Cavalli-Sforza, Samhaa R El-Beltagy, Wassim El-Hajj, and 1 others. 2021. A panoramic survey of natural language processing in the arab world. Communications of the ACM, 64(4):72–81. find
  20. Passant Elchafei and Amany Fashwan. 2025. Multimodal arabic captioning with interpretable visual concept integration. arXiv preprint arXiv:2510.03295. ↗
  21. Rouzbeh Ghasemi and Saeedeh Momtazi. 2023. How a deep contextualized representation and attention mechanism justifies explainable cross-lingual sentiment analysis. ACM Transactions on Asian and Low-Resource Language Information Processing, 22(11):1–15. ↗
  22. Nizar Y Habash. 2010. Introduction to Arabic natural language processing. Morgan & Claypool Publishers. ↗
  23. Md Mehedi Hassan, Anindya Nag, Riya Biswas, Md Shahin Ali, Sadika Zaman, Anupam Kumar Bairagi, and Chetna Kaushal. 2025. Explainable artificial intelligence for natural language processing: A survey. Data & Knowledge Engineering, 160:102470. ↗
  24. Md Mithun Hossain, Md Shakil Hossain, Mejdl Safran, Sultan Alfarhood, Meshal Alfarhood, and Muhammad Firoz Mridha. 2024. A hybrid attention-based transformer model for arabic news classification using text embedding and deep learning. IEEE Access, 12:198046–198066. ↗
  25. Huang Huang, Fei Yu, Jianqing Zhu, Xuening Sun, Hao Cheng, Song Dingjie, Zhihong Chen, Mosen Alharthi, Bang An, Juncai He, and 1 others. 2024. Acegpt, localizing large language models in arabic. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), pages 8139–8163.
  26. Omar Ibrahim Aboulola and Muhammad Umer. 2024. Novel approach for arabic fake news classification using embedding from large language features with cnn-lstm ensemble model and explainable ai. Scientific Reports, 14(1):30463. ↗
  27. Go Inoue, Bashar Alhafni, Nurpeiis Baimukan, Houda Bouamor, and Nizar Habash. 2021. The interplay of variant, size, and task type in arabic pre-trained language models. In Proceedings of the sixth Arabic natural language processing workshop, pages 92– 104.
  28. Ferihane Kboubi and Anja Habacha Chaibi. 2025. Fine-grained arabic dialect identification: Investigating various approaches across multiple datasets. ACM Transactions on Asian and Low-Resource Language Information Processing, 24(10):1–33. ↗
  29. Khalid A. Khader, Mohammed S. Hussein, and Abdallatif S. Abu-Issa. 2025. Adapting large language models for arabic: Comparative evaluation, fine-tuning, and ethical deployment. IEEE Access. ↗
  30. Mohamed Bayan Kmainasi, Abul Hasnat, Md Arid Hasan, Ali Ezzat Shahroor, and Firoj Alam. 2025. Memeintel: Explainable detection of propagandistic and hateful memes. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 30263–30279. find
  31. Abhinav Kumar, Jyoti Kumari, and Jiesth Pradhan. 2023. Explainable deep learning for mental health detection from english and arabic social media posts. ACM Transactions on Asian and Low-Resource Language Information Processing. ↗
  32. Siwen Luo, Hamish Ivison, Soyeon Caren Han, and Josiah Poon. 2024. Local interpretations for explainable natural language processing: A survey. ACM Computing Surveys, 56(9):1–36. ↗
  33. Rim Mahouachi. 2026. A hybrid multi-view and neuro-symbolic approach for arabic offensive language detection: A step towards safer online spaces. Expert Systems with Applications, page 131854. ↗
  34. Ahmad M. Mustafa, Saja Nakhleh, Rama Irsheidat, and Raneem Alruosan. 2024. Interpreting arabic transformer models: A study on xai interpretability for qur’anic semantic-search models. Jordanian Journal of Computers and Information Technology, 10. ↗
  35. Nour Rabih. 2025. Interpretable measures for arabic readability. ↗
  36. Abdulwahab Sahyoun and Shady Shehata. 2023. Aradiawer: An explainable metric for dialectical arabic asr. In Proceedings of the Second Workshop on NLP Applications to Field Linguistics, pages 64–73. ↗
  37. Neha Sengupta, Sunil Kumar Sahu, Bokang Jia, Satheesh Katipomu, Haonan Li, Fajri Koto, William Marshall, Gurpreet Gosal, Cynthia Liu, Zhiming Chen, and 1 others. 2023. Jais and jais-chat: Arabic-centric foundation and instruction-tuned open generative large language models. arXiv preprint arXiv:2308.16149. ↗
  38. Zhiqiang Shi and Ruchit Agrawal. 2025. A comprehensive survey of contemporary arabic sentiment analysis: Methods, challenges, and future directions. Findings of the Association for Computational Linguistics: NAACL 2025, pages 3760–3772.
  39. Asmaa Hashem Sweidan, Nashwa El-Bendary, Shereen A. Taie, Amira M. Idrees, and Esraa Elhariri. 2025. Explainable deep learning for covid-19 vaccine sentiment in arabic tweets using multi-self-attention bilstm with xlnet. Big Data and Cognitive Computing, 9(2):37. ↗
  40. Ahmed Mustafa Younes. 2026. Deformar: A visual analytics framework for evaluation of arabic named entity recognition. In Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script, pages 253–275. ↗
  41. Mohamad Bilal Zbib, Hasan Abed Al Kader Hammoud, Ammar Mohanna, Nadine Rizk, Fatima Karnib, Sina Moukaled, and Bernard Ghanem. 2026. Aralingbench: A human-annotated benchmark for evaluating arabic linguistic capabilities of large language models. In Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script, pages 385–393.
  42. Haiyan Zhao, Hanjie Chen, Fan Yang, Ninghao Liu, Huiqi Deng, Hengyi Cai, Shuaiqiang Wang, Dawei Yin, and Mengnan Du. 2024. Explainability for large language models: A survey. ACM Transactions on Intelligent Systems and Technology, 15(2):1–38. ↗
  43. Julia El Zini and Mariette Awad. 2022. On the explainability of natural language processing deep models. ACM Computing Surveys, 55(5):1–31. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/lamsiyah-2026-why-current-xai-enough-arabic/