Elamur Research⌕
☰
home / papers / 2025 / Integrating Argumentation Features for Enhanced Propaganda Detection in Arabic Narratives on the Israeli War on Gaza

Integrating Argumentation Features for Enhanced Propaganda Detection in Arabic Narratives on the Israeli War on Gaza

Paper · 2025 · Sara Nabhani, Claudia Borg, Kurt Micallef et al.

Summary

The paper asks whether knowing the argumentative structure of a text helps a model find propaganda techniques in Arabic. Earlier Arabic systems mostly fine-tuned transformers or added data augmentation. The authors instead give the model explicit argument components: Major Claim, Claim and Premise. They train on the ArAIEval 2024 Task 1 data, which contains tweets and news paragraphs annotated with 23 fine-grained persuasion techniques as overlapping, multi-label spans. No suitable Arabic argument-mining corpus exists, so they take argumentation knowledge from the English Persuasive Essays corpus. They move it to Arabic with cross-lingual and multilingual methods and add argument labels produced by GPT-4. The resulting argument features improve over the baselines consistently. The authors then run the enhanced model on Arabic items from the FIGNEWS corpus on the Israeli war on Gaza and analyse the output by hand. The model catches many rhetorical strategies but shows clear error patterns. The work matters because it shows that discourse-level structure can strengthen detection of manipulation in a low-resource, politically sensitive setting.

Key points

  • Propaganda data: ArAIEval 2024 Task 1, with the official splits used unchanged.
    • Train: 6,997 documents.
    • Dev: 921 documents.
    • Test: 1,046 documents.
    • The 23 techniques follow the Piskorski et al. (2023) taxonomy, and Loaded Language and Name Calling dominate.
  • Argument data: the Persuasive Essays corpus (Stab and Gurevych, 2017).
    • Size: 402 essays, split 322 for training and 80 for testing.
    • Labels: 751 MajorClaim, 1,506 Claim and 3,832 Premise.
    • It was used because no fitting Arabic argument dataset was available.
  • Transfer to Arabic relied on cross-lingual and multilingual modelling together with GPT-4-based annotation. The resulting features gave steady gains over the baselines.
  • Qualitative study: 17 Arabic examples from FIGNEWS (Zaghouani et al., 2024), sampled from the top shared-task system of team NLPColab. FIGNEWS also carries bias and propaganda labels.
  • Error types found in the qualitative study:
    • over-prediction of Loaded Language on neutral words such as "We warned";
    • spans that are too wide, for example ones that include a preposition;
    • missed cases, such as a single-solution framing that could be labelled False Dilemma or Appeal to Fear.
  • Code is published at github.com/saranabhani/prop-arg.

Interesting ideas

  • I1. Argument roles (claim, premise, major claim) are added as side features for a different target task. The intuition is that propaganda reuses the shape of an argument but loads it with emotion, so the structure helps separate persuasion from manipulation.
  • I2. When the target language lacks a resource, the authors borrow argument annotation from an English corpus and add silver labels from GPT-4 instead of annotating from scratch.
  • I3. The label scheme lets one character span carry several techniques and lets spans overlap. This avoids forcing a single tag where real rhetoric stacks devices.
  • I4. The hand-made error taxonomy (over-prediction, overly broad spans, missed content) shows where a span model goes wrong in ways that a single F1 score hides.

For Mova

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

  • I1 (frames): prag could tag clauses as claim or premise, then flag a claim with no supporting premise plus emotive wording as a manipulative speech act. This gives an explainable cue for implicature and irony.
  • I2 (learn): this matches the LLM teacher tiers. An LLM could pre-label rare discourse categories that witness parsers then check, giving silver data where no gold treebank exists.
  • I3 (frames): frames in prag could likewise hold several pragmatic labels per span, for example a phrase that is both exaggeration and name calling, instead of one exclusive tag.
  • I4 (tree): span boundaries could be snapped to UD subtrees, dropping a leading case/ADP dependent like "Over". Separately, Mova's evaluation could count these three error types.

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 GroningenGroningen · Netherlands74 works · 5,142 citations🎓University of MaltaMsida · Malta32 works · 895 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksSocial Sciences · 1,954 worksSociology and Political Science · 626 works

🧭 Topics 1Research subjects and languages the work is about.Semitic languages · 3,066 works

Cite

Cite as Sara Nabhani, Claudia Borg, Kurt Micallef and Khalid Al-Khatib. 2025. Integrating Argumentation Features for Enhanced Propaganda Detection in Arabic Narratives on the Israeli War on Gaza. Proceedings of the first International Workshop on Nakba Narratives as Language Resources.

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

@inproceedings{nabhani2025integrating,
  title = {{Integrating Argumentation Features for Enhanced Propaganda Detection in Arabic Narratives on the Israeli War on Gaza}},
  author = {Sara Nabhani, Claudia Borg, Kurt Micallef and Khalid Al-Khatib},
  booktitle = {Proceedings of the first International Workshop on Nakba Narratives as Language Resources},
  year = {2025},
  url = {https://aclanthology.org/2025.nakbanlp-1.14/}
}

References

26 references, 9 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Muhammad AbdulMageed, AbdelRahim Elmadany, and El Moatez Billah Nagoudi. 2021. ARBERT & MARBERT: Deep bidirectional transformers for Ara bic. In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Lan guage Processing (Volume 1: Long Papers), pages 7088–7105, Online. Association for Computational Linguistics. find
  2. Sadaf Abdul Rauf, Huda Sarfraz, Saadia Nauman, Arooj Fatima, SadafZiafat SadafZiafat, Momina Ishfaq, Al ishba Suboor, Hammad Afzal, and Seemab Latif. 2024. NLPColab at FigNews 2024 shared task: Chal lenges in bias and propaganda annotation for news media. In Proceedings of The Second Arabic Natural Language Processing Conference, pages 573–579, Bangkok, Thailand. Association for Computational Linguistics. find
  3. Firoj Alam, Hamdy Mubarak, Wajdi Zaghouani, Giovanni Da San Martino, and Preslav Nakov. 2022. Overview of the wanlp 2022 shared task on propaganda detection in arabic. Preprint, arXiv:2211.10057. ↗
  4. Khudejah Ali and Khawaja Zain ul abdin. 2021. Post truth propaganda: heuristic processing of political fake news on facebook during the 2016 u.s. presi dential election. Journal of Applied Communication Research, 49(1):109–128. find
  5. Wissam Antoun, Fady Baly, and Hazem Hajj. 2020. AraBERT: Transformerbased model for Arabic lan guage understanding. In Proceedings of the 4th Work shop on OpenSource Arabic Corpora and Process ing Tools, with a Shared Task on Offensive Language Detection, pages 9–15, Marseille, France. European Language Resource Association. find
  6. Mikel Artetxe, Vedanuj Goswami, Shruti Bhosale, An gela Fan, and Luke Zettlemoyer. 2023. Revisiting machine translation for crosslingual classification. In Proceedings of the 2023 Conference on Empiri cal Methods in Natural Language Processing, pages 6489–6499, Singapore. Association for Computa tional Linguistics. find
  7. Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettle moyer, and Veselin Stoyanov. 2019. Unsupervised crosslingual representation learning at scale. CoRR, abs/1911.02116. ↗
  8. Jacob Devlin, MingWei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pretraining of deep bidirectional transformers for language under standing. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Tech nologies, Volume 1 (Long and Short Papers), pages 4171–4186. find
  9. Chris Dyer, Victor Chahuneau, and Noah A. Smith. 2013. A simple, fast, and effective reparameterization of IBM model 2. In Proceedings of the 2013 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Tech nologies, pages 644–648, Atlanta, Georgia. Associa tion for Computational Linguistics. find
  10. Maram Hasanain, Firoj Alam, Hamdy Mubarak, Samir Abdaljalil, Wajdi Zaghouani, Preslav Nakov, Gio vanni Da San Martino, and Abed Freihat. 2023. ArAIEval shared task: Persuasion techniques and disinformation detection in Arabic text. In Proceed ings of ArabicNLP 2023, pages 483–493, Singapore (Hybrid). Association for Computational Linguistics.
  11. Maram Hasanain, Md. Arid Hasan, Fatema Ahmed, Reem Suwaileh, Md. Rafiul Biswas, Wajdi Za ghouani, and Firoj Alam. 2024. Araieval shared task: Propagandistic techniques detection in uni modal and multimodal arabic content. Preprint, arXiv:2407.04247. ↗
  12. Ahmed Samir Hussein, Abu Bakr Soliman Moham mad, Mohamed Ibrahim, Laila Hesham Afify, and Samhaa R. ElBeltagy. 2022. NGU CNLP atWANLP 2022 shared task: Propaganda detection in Arabic. In Proceedings of the The Seventh Arabic Natural Language Processing Workshop (WANLP), pages 545–550, Abu Dhabi, United Arab Emirates (Hybrid). Association for Computational Linguistics.
  13. David Broniatowski, Daniel Kerchner, Fouzia Farooq, Xiaolei Huang, Amelia Jamison, Mark Dredze, and Sandra Crouse Quinn. 2020. The covid19 social me dia infodemic reflects uncertainty and statesponsored propaganda. arXiv preprint. find
  14. Momtazul Labib, Samia Rahman, Hasan Murad, and Udoy Das. 2024. CUET_sstm at ArAIEval shared task: Unimodal (text) propagandistic technique detec tion using transformerbased model. In Proceedings of The Second Arabic Natural Language Processing Conference, pages 507–511, Bangkok, Thailand. As sociation for Computational Linguistics. find
  15. Alexander Chernyavskiy, Dmitry Ilvovsky, and Preslav Nakov. 2024. Unleashing the power of discourse enhanced transformers for propaganda detection. In Salima Lamsiyah, Abdelkader El Mahdaouy, Hamza Alami, Ismail Berrada, and Christoph Schommer. 2023. UL & UM6P at ArAIEval shared task: Transformerbased model for persuasion techniques and disinformation detection in Arabic. In Proceed ings of ArabicNLP 2023, pages 558–564, Singapore (Hybrid). Association for Computational Linguistics. find
  16. Yinhan Liu, Myle Ott, Naman Goyal, Jingfei Du, Man dar Joshi, Danqi Chen, Omer Levy, Mike Lewis, Luke Zettlemoyer, and Veselin Stoyanov. 2019. Roberta: A robustly optimized bert pretraining ap proach. Preprint, arXiv:1907.11692. ↗
  17. Ana Laura Nettel and Georges Roque. 2012. Persuasive argumentation versus manipulation. Argumentation, 26(1):55–69. ↗
  18. Yizhong Wang, Sujian Li, and Houfeng Wang. 2017. A twostage parsing method for textlevel discourse analysis. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Vol ume 2: Short Papers), pages 184–188, Vancouver, Canada. Association for Computational Linguistics. find
  19. Wajdi Zaghouani, Mustafa Jarrar, Nizar Habash, Houda Bouamor, Imed Zitouni, Mona Diab, Samhaa R. ElBeltagy, and Muhammed AbuOdeh. 2024. The fignews shared task on news media narratives. Preprint, arXiv:2407.18147. ↗
  20. Jakub Piskorski, Nicolas Stefanovitch, Giovanni Da San Martino, and Preslav Nakov. 2023. SemEval 2023 task 3: Detecting the category, the framing, and the persuasion techniques in online news in a multi lingual setup. In Proceedings of the 17th Interna tional Workshop on Semantic Evaluation (SemEval 2023), pages 2343–2361, Toronto, Canada. Associa tion for Computational Linguistics.
  21. Md Riyadh and Sara Nabhani. 2024. Mela at ArAIEval shared task: Propagandistic techniques detection in Arabic with a multilingual approach. In Proceedings of The Second Arabic Natural Language Processing Conference, pages 478–482, Bangkok, Thailand. As sociation for Computational Linguistics. ↗
  22. Ali Safaya, Moutasem Abdullatif, and Deniz Yuret. 2020. KUISAIL at SemEval2020 task 12: BERTCNN for offensive speech identification in social media. In Proceedings of the Fourteenth Workshop on Seman tic Evaluation, pages 2054–2059, Barcelona (online). International Committee for Computational Linguis tics. find
  23. Zien Sheikh Ali, Watheq Mansour, Tamer Elsayed, and Abdulaziz Al‐Ali. 2021. AraFacts: The first large Arabic dataset of naturally occurring claims. In Pro ceedings of the Sixth Arabic Natural Language Pro cessing Workshop, pages 231–236, Kyiv, Ukraine (Virtual). Association for Computational Linguistics.
  24. Christian Stab and Iryna Gurevych. 2017. Parsing argu mentation structures in persuasive essays. Computa tional Linguistics, 43(3):619–659. find
  25. NLLB Team, Marta R. Costajussà, James Cross, Onur Çelebi, Maha Elbayad, Kenneth Heafield, Kevin Hef fernan, Elahe Kalbassi, Janice Lam, Daniel Licht, Jean Maillard, Anna Sun, Skyler Wang, Guillaume Wenzek, Al Youngblood, Bapi Akula, Loic Bar rault, Gabriel Mejia Gonzalez, Prangthip Hansanti, John Hoffman, Semarley Jarrett, Kaushik Ram Sadagopan, Dirk Rowe, Shannon Spruit, Chau Tran, Pierre Andrews, Necip Fazil Ayan, Shruti Bhosale, Sergey Edunov, Angela Fan, Cynthia Gao, Vedanuj Goswami, Francisco Guzmán, Philipp Koehn, Alexandre Mourachko, Christophe Rop ers, Safiyyah Saleem, Holger Schwenk, and Jeff Wang. 2022. No language left behind: Scal ing humancentered machine translation. Preprint, arXiv:2207.04672. ↗
  26. A Propaganda Techniques Definition In this section, we provide the definitions of the propaganda techniques included in the dataset, as outlined in (Piskorski et al., 2023). find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/nabhani-2025-integrating-argumentation-features-enhanced-propaganda/