Elamur Research⌕
☰
home / papers / 2025 / A Novel Dialect-Aware Framework for the Classification of Arabic Dialects and Emotions

A Novel Dialect-Aware Framework for the Classification of Arabic Dialects and Emotions

Summary

The paper addresses two linked tasks for Arabic text: identifying which regional dialect a text is written in, and detecting the emotion it expresses while accounting for dialect. Earlier Arabic emotion work mostly ignored dialect variation and relied on scarce annotated data. The author proposes a framework with three modules. A preprocessing module prepares the text. A classification module, built on AraBERT variants and compared against traditional classifiers, predicts dialect and emotion. A clustering module induces new emotion lexicons that are specific to each dialect, starting from a corpus labelled with dialects. AraBERT-large reaches 88.9% accuracy on four-way AOC dialect classification (MSA, GLF, EGY, LEV), which is 6.45 points above a prior BiLSTM result. Emotion classifiers trained per dialect beat a pooled general classifier by about 4–5 points for Egyptian and Gulf Arabic. Native speakers checked the induced lexicons. The main value is the procedure for generating emotion lexicons per dialect, which needs only native-speaker verification rather than linguist annotation.

Key points

  • Dialect classification (Table 3): AraBERT-large 89.2%, AraBERT-base 87.2%, AraBERT-Twitter 82.6%, Multinomial Naive Bayes 86.5%, Random Forest 76.0%, Linear SVM 74.7%, versus 69.2% reported by Alshutayri.
  • On AOC (MSA/GLF/EGY/LEV), AraBERT-large scores 88.9% against 82.45% for the BiLSTM of Elaraby and Abdul-Mageed.
  • Dialect-aware emotion classification: EGY 89.1% vs. 84.2% for the general model; GLF 79% vs. 75.1%.
  • The six Ekman-style emotions are anger, disgust, fear, joy, sadness and surprise. Surprise is the hardest class (precision 0.6 GLF, 0.5 EGY), and sadness is also weak for GLF (0.79).
  • Lexicon validity was checked only for EGY and GLF, by three native speakers per dialect, because only those speakers were available to the author.

Interesting ideas

  • I1. Train a model separately for each language variety and compare it directly against a single pooled model on the same data, so the gain from conditioning on variety becomes a measured number rather than an assumption. — For Mova (learn): Use this as a standard ablation whenever a component is split by genre or English dialect, with the pooled model serving as the control that the split must beat.
  • I2. Induce lexicon entries automatically from labelled data per subgroup, then have ordinary native speakers verify them instead of specialist linguists, which keeps lexicon expansion cheap. — For Mova (frames): Emotion and evaluative lexicons used for irony and speech-act cues could be grown from text clusters and accepted only after lightweight human or teacher checks, each entry keeping its provenance for explainability.
  • I3. Treat dialect identification as an upstream step whose output selects the downstream emotion model, rather than folding everything into one classifier. — For Mova (tree): Detecting the English variety or register first (e.g. social-media vs. edited text) could route input to a treebank-specific parser or conversion rules, with the routing decision logged as part of the explanation.
  • I4. A simple word-voting baseline counts how many tokens belong to each dialect's vocabulary, giving a transparent reference point for heavier models. — For Mova (tokens): Lexicon-vote features per variety could flag spelling or vocabulary variants before tagging, offering an interpretable signal that is easy to audit.
  • I5. Validation was restricted to the dialects where native checkers existed, and the paper states this limit openly instead of overclaiming coverage. — For Mova (answer): Report evaluation scope explicitly, stating which text types were actually human-checked, so answer-quality claims stay tied to verified domains.

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

Cited by

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

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.3844/jcssp.2025.88.95 · arXiv 2502.09128

@misc{alsadhan2025novel,
  title = {{A Novel Dialect-Aware Framework for the Classification of Arabic Dialects and Emotions}},
  author = {Alsadhan, Nasser A},
  year = {2025},
  doi = {10.3844/jcssp.2025.88.95},
  eprint = {2502.09128},
  archivePrefix = {arXiv},
  url = {https://doi.org/10.3844/jcssp.2025.88.95}
}

References

28 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Abdullah, M., AlMasawa, M., Makki, I., Alsolmi, M., & Mahrous, S. (2020). Emotions Extraction from Arabic Tweets. International Journal of Computers and Applications, 42(7), 661–675. https://doi.org/10.1080/1206212x.2018.1482395 ↗
  2. Nasser Alsadhan / Journal of Computer Science 2025, 21 (1): 88.95 DOI: 10.3844/jcssp.2025.88.95 ↗
  3. Aljuhani, R. H., Alshutayri, A., & Alahdal, S. (2021). Arabic Speech Emotion Recognition from Saudi Dialect Corpus. IEEE Access, 9, 127081–127085. https://doi.org/10.1109/access.2021.3110992 ↗
  4. Al-Khatib, A., & El-Beltagy, S. R. (2018). Emotional Tone Detection in Arabic Tweets (Vol. 10762, pp. 105–114). Springer International Publishing. https://doi.org/10.1007/978-3-319-77116-8_8 ↗
  5. Alsadhan, N., & Skillicorn, D. (2017). Estimating Personality from Social Media Posts. 2017 IEEE International Conference on Data Mining Workshops (ICDMW), 350–356. https://doi.org/10.1109/icdmw.2017.51 ↗
  6. Alshutayri, A. (2018). Arabic Dialect Texts Classification. find
  7. Alsarsour, I., Mohamed, E., Suwaileh, R., & Elsayed, T. (2018). Dart: A large Dataset of Dialectal Arabic Tweets. Proceedings of the Eleventh International Conference on Language Resources and Evaluation, 3666–3670. find
  8. Alshutayri, A., & Atwell, E. (2019). Classifying Arbic Dialect Text in the Social Media Arabic Dialect Corpus (SMADC). Proceedings of the 3 rd Workshop on Arabic Corpus Linguistics, 51–59. find
  9. Alsudais, A., Alotaibi, W., & Alomary, F. (2022). Similarities Between Arabic dialects: Investigating Geographical Proximity. Information Processing & Management, 59(1), 102770–102770. https://doi.org/10.1016/j.ipm.2021.102770 ↗ our card
  10. Antoun, W., Baly, F., & Hajj, H. (2020). Transformer-based Model for Arabic Language Understanding,”. ArXiv, arXiv:2003.00104. ↗ our card
  11. Biadsy, F. (2011). Automatic Dialect and Accent Recognition and its Application to Speech Recognition. https://doi.org/10.7916/D8M61S68 ↗
  12. Bojanowski, P., Grave, E., Joulin, A., & Mikolov, T. (2017). Enriching Word Vectors with Subword Information. Transactions of the Association for Computational Linguistics, 5, 135–146. https://doi.org/10.1162/tacl_a_00051 ↗
  13. Cotterell, R., & Callison-Burch, C. (2014). A Multi-Dialect, Multi-Genre Corpus of Informal Written Arabic. In LREC, 241–245. find
  14. Dahmani, H., Hussein, H., Meyer-Sickendiek, B., & Jokisch, O. (2019). Natural Arabic Language Resources for Emotion Recognition in Algerian Dialect (K. Smaïli, Ed.; Vol. 1108, pp. 18–33). Springer International Publishing. https://doi.org/10.1007/978-3-030-32959-4_2 ↗
  15. Elaraby, M., & Abdul-Mageed, M. (2018). Deep models for Arabic dialect identification on benchmarked data. Proceedings of the Fifth Workshop on NLP for Similar Languages, Varieties and Dialects, 263-274. find
  16. Ekman, P. (1992). An Argument for Basic Emotions. Cognition and Emotion, 6(3–4), 169–200. https://doi.org/10.1080/02699939208411068 ↗ our card
  17. Ester, M., Kriegel, H. P., Sander, J., & Xu, X. (1996). A density-based algorithm for discovering clusters in large spatial databases with noise. KDD, 226–231. find
  18. El-Haj, M., Rayson, P., & Aboelezz, M. (2018). Arabic Dialect Identification in the Context of Bivalency and Code-Switching. European Language Resources Association, 3622–3627. find
  19. Habash, N. Y. (2010). Introduction to Arabic Natural Language Processing (1st ed., Vol. 3). Springer. https://doi.org/10.1007/978-3-031-02139-8 ↗
  20. Lulu, L., & Elnagar, A. (2018). Automatic Arabic Dialect Classification Using Deep Learning Models. Procedia Computer Science, 142(Procedia Computer Science), 262–269. https://doi.org/10.1016/j.procs.2018.10.489 ↗
  21. Moudjari, L., Akli-Astouati, K., & Benamara, F. (2020). An Algerian corpus and an annotation platform for opinion and emotion analysis. Proc. LREC, 1202-1210. find
  22. Eslam., Mosa, M., & Ismail, N. (2019). Emotion Analysis in Arabic Language Applying Transfer Learning. 2019 15 th International Computer Engineering Conference (ICENCO), 204–209. https://doi.org/10.1109/icenco48310.2019.9027295 ↗
  23. Rabab’ah, A. M., Al-Ayyoub, M., Jararweh, Y., & Al-Kabi, M. N. (2016). Evaluating SentiStrength for Arabic Sentiment Analysis. 2016 7 th International Conference on Computer Science and Information Technology (CSIT), 1–6. https://doi.org/10.1109/csit.2016.7549458 ↗
  24. Saad, M. (2015). Mining Documents and Sentiments in Cross-Lingual Context. Université de Lorraine. find
  25. Sharmila, D., Kannimuthu, S., Ravikumar, G., & Anand, K. (2019). KCE DALab-APDA@ FIRE2019: Author Profiling and Deception Detection in Arabic Using Weighted Embedding. CEUR Workshop Proceedings, 12–15. find
  26. Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., & Jones, L. (2017). Attention is All you Need. Advances in Neural Information Processing Systems, 6000–6010. find
  27. Wolf, T., Debut, L., Sanh, V., Chaumond, J., Delangue, C., Moi, A., Cistac, P., Rault, T., Louf, R., Funtowicz, M., Davison, J., Shleifer, S., von Platen, P., Ma, C., Jernite, Y., Plu, J., Xu, C., Le Scao, T., Gugger, S., … Rush, A. (2020). Transformers: State-of-the-Art Natural Language Processing. Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, 38–45. https://doi.org/10.18653/v1/2020.emnlp-demos.6 ↗ our card
  28. Zaidan, O., & Callison-Burch, C. (2011). The Arabic Online Commentary Dataset: An Annotated Dataset of Informal Arabic with High Dialectal Content. Proceedings of the 49 th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies, 37–41. find our card

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/alsadhan-2025-novel-dialect-aware-framework-classification/