---
title: "Transfer or Translate? Argument Mining in Arabic with No Native"
description: "The study offers a reusable recipe and baseline for low-resource argument mining, and the resources are released on GitHub. 16 references."
authors: ["Sara Nabhani", "Khalid Al Khatib"]
year: 2025
venue: "Proceedings of The Third Arabic Natural Language Processing Conference"
source: "https://aclanthology.org/2025.arabicnlp-main.33/"
doi: "10.18653/v1/2025.arabicnlp-main.33"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/nabhani-2025-transfer-or-translate-argument-mining/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/nabhani-2025-transfer-or-translate-argument-mining/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2025](https://md.research.elamur.ai/free/en/papers/2025/page/19/) / Transfer or Translate? Argument Mining in Arabic with No Native Annotations

# Transfer or Translate? Argument Mining in Arabic with No Native Annotations

**Authors:** [Sara Nabhani](https://md.research.elamur.ai/free/en/authors/sara-nabhani/) and [Khalid Al Khatib](https://md.research.elamur.ai/free/en/authors/khalid-al-khatib/) · **Year:** 2025 · **Venue:** [Proceedings of The Third Arabic Natural Language Processing Conference](https://md.research.elamur.ai/free/en/venues/arabicnlp/) ([ArabicNLP 2025](https://md.research.elamur.ai/free/en/venues/arabicnlp/2025/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2025.arabicnlp-main.33/) · [PDF](https://aclanthology.org/2025.arabicnlp-main.33.pdf) · [DOI 10.18653/v1/2025.arabicnlp-main.33](https://doi.org/10.18653/v1/2025.arabicnlp-main.33) · [OpenAlex](https://openalex.org/W4416037673)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
Arabic argument mining has almost no annotated corpora that mark argument structure. Nabhani and Al Khatib ask whether English annotations can stand in for native Arabic ones. They treat argument component identification as span-level BIO sequence labeling over paragraphs from the English Persuasive Essays (PE) corpus. The labels are Major Claim, Claim and Premise. They compare five setups:
- a multilingual model trained on English and applied to Arabic zero-shot;
- multilingual models trained on machine-translated Arabic, alone or mixed with English;
- a monolingual Arabic model trained on the translated data;
- zero-shot prompting of LLMs.

The training data was translated with NLLB, and span labels were carried over with FastAlign word alignment. All systems are scored on a manually corrected Arabic test set. The monolingual Arabic model trained on translated data performs best. Hand-fixing a modest share of the projected training labels improves it further. Absolute scores stay low, though, so the authors argue that a native, human-annotated Arabic corpus is still required. The study offers a reusable recipe and baseline for low-resource argument mining, and the resources are released on GitHub.

## Key points
- **Source corpus:** PE (Stab and Gurevych, 2017).
  - 402 essays split 322 train / 80 test, giving 2,235 paragraphs and 148,182 tokens.
  - Totals are 751 Major Claims, 1,506 Claims and 3,832 Premises.
  - Each paragraph is one instance.
- **Pipeline and models:**
  - Translation uses NLLB, and label projection uses FastAlign.
  - Multilingual runs use XLM-RoBERTa-large; the Arabic monolingual model is AraBERT.
  - The LLMs, prompted zero-shot, are Llama 3.1 70B and Claude 3.5 Haiku.
- **Manual correction:** reviewing 400 projected training instances (under 23% of the training set) raised AraBERT F1 from 0.251 to 0.351.
- **Ranking:** translate-train monolingual beat all multilingual variants. Zero-shot transfer and LLM prompting were clearly weaker.
- **Alignment limits:** FastAlign struggles with long argument spans and with loose, non-literal translations, which makes label noise the main bottleneck.

## Interesting ideas
- Hand-correcting a limited slice of automatically projected training labels yields a large, measurable gain, which shows where human effort pays off most in a noisy pipeline.
- Training data can stay noisy while the test set is cleaned by hand, so gains are judged against trustworthy gold rather than the same projection errors.
- When spans are carried across a translation through word alignment, long and non-literal spans are where projection breaks. Error audits should therefore target span length and translation literalness.
- Claims, premises and major claims can be modeled as BIO spans at paragraph level. This gives a simple, inspectable discourse layer above sentence parsing.
- A specialized single-language model trained on imperfect in-language data beat a broad multilingual model, suggesting that matching the target language matters more than cross-lingual breadth.

## For Mova

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

- **I1** (learn): Disagreement cases in the silver data from witness parsers could be ranked, and a small batch hand-fixed. The resulting jump would show how much silver noise limits the parser.
- **I2** (learn): Any metric reported on silver data should be paired with a small manually verified evaluation set. Otherwise systematic silver errors look like correct answers.
- **I3** (back): When translating through the meta-world, check whether long multi-clause units keep their role boundaries. These boundaries are the likely failure point for any alignment-based mapping.
- **I4** (frames): prag could add an argument-role layer that tags clauses as claim or premise from frames and connectives. This would let readers explain why a text concludes something.
- **I5** (tree): When choosing gparse training data, English-specific (even silver) treebanks may outperform pooled multilingual resources for an English-only parser.

*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 Groningen](https://md.research.elamur.ai/free/en/universities/university-of-groningen/) — University, Groningen, Netherlands · 74 works · 5,142 citations

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 17,450 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 17,296 works

🧭 Topics (1): [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 3,066 works

## Cite

Cite as: Sara Nabhani and Khalid Al Khatib. 2025. Transfer or Translate? Argument Mining in Arabic with No Native Annotations. Proceedings of The Third Arabic Natural Language Processing Conference.

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

```bibtex
@inproceedings{nabhani2025transfer,
  title = {{Transfer or Translate? Argument Mining in Arabic with No Native Annotations}},
  author = {Sara Nabhani and Khalid {Al Khatib}},
  booktitle = {Proceedings of The Third Arabic Natural Language Processing Conference},
  year = {2025},
  doi = {10.18653/v1/2025.arabicnlp-main.33},
  url = {https://doi.org/10.18653/v1/2025.arabicnlp-main.33}
}
```

- [← previous Tracing Syntactic Complexity: Exploring the Evolution of Average Dependency Length Across Three Centuries of Scientific English](https://md.research.elamur.ai/free/en/papers/krielke-2025-tracing-syntactic-complexity-exploring-evolution/)
- [next → Two-Stage Quranic QA via Ensemble Retrieval and Instruction-Tuned Answer Extraction](https://md.research.elamur.ai/free/en/papers/basem-2025-two-stage-quranic-qa-ensemble/)
- [↑ back to papers of 2025](https://md.research.elamur.ai/free/en/papers/2025/page/19/)

## References

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

1. Abdul Gabbar Al-Sharafi, Mohammad Majed Khader, Mohamed Ahmed, Mohamad Hamza Al-Sioufy, Wajdi Zaghouani, and Ali Al-Zawqari. 2025. A hybrid annotation model for arabic argumentative debate corpus. In Arabic Language Processing, Communications in Computer and Information Science, pages 97– 113, Germany. Springer Science and Business Media Deutschland GmbH. Publisher Copyright: © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.; 8th International Conference on Arabic Language Processing, ICALP 2023 ; Conference date: 19-04-2024 Through 20-04-2024. [find](https://scholar.google.com/scholar?q=Abdul+Gabbar+Al%2DSharafi%2C+Mohammad+Majed+Khader%2C+Mohamed+Ahmed%2C+Mohamad+Hamza+Al%2DSioufy%2C+Wajdi+Zaghouani%2C+and+Ali+Al%2DZawqari%2E+2025%2E+A+hybrid+annotation+model+for+arabic+argumentative+debate+corpus%2E+In+)
2. Wissam Antoun, Fady Baly, and Hazem Hajj. [Arabert: Transformer-based model for arabic language understanding](https://md.research.elamur.ai/free/en/papers/antoun-2020-arabert-transformer-based-model-arabic/). In LREC 2020 Workshop Language Resources and Evaluation Conference 11–16 May 2020, page 9.
3. Elena Cabrio and Serena Villata. 2018. Five years of argument mining: A data-driven analysis. In Proceedings of the 27th International Joint Conference on Artificial Intelligence, pages 5427–5433. ACM. [find](https://scholar.google.com/scholar?q=Elena+Cabrio+and+Serena+Villata%2E+2018%2E+Five+years+of+argument+mining%3A+A+data%2Ddriven+analysis%2E+In+Proceedings+of+the+27th+International+Joint+Conference+on+Artificial+Intelligence%2C+pages+5427%E2%80%935433%2E+ACM)
4. Guizhen Chen, Liying Cheng, Anh Tuan Luu, and Lidong Bing. 2024. Exploring the potential of large language models in computational argumentation. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 2309–2330, Bangkok, Thailand. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Guizhen+Chen%2C+Liying+Cheng%2C+Anh+Tuan+Luu%2C+and+Lidong+Bing%2E+2024%2E+Exploring+the+potential+of+large+language+models+in+computational+argumentation%2E+In+Proceedings+of+the+62nd+Annual+Meeting+of+the+Assoc)
5. Alexis Conneau, Kartikay Khandelwal, Naman Goyal, Vishrav Chaudhary, Guillaume Wenzek, Francisco Guzmán, Edouard Grave, Myle Ott, Luke Zettlemoyer, and Veselin Stoyanov. 2019. [Unsupervised cross-lingual representation learning at scale](https://md.research.elamur.ai/free/en/papers/conneau-2020-unsupervised-cross-lingual-representation-learning/). CoRR, abs/1911.02116. [↗](https://arxiv.org/abs/1911.02116)
6. Marta R. Costa-jussà, James Cross, Onur Çelebi, Maha Elbayad, Kenneth Heafield, Kevin Heffernan, and Elahe Kalbassi et al. 2022. No language left behind: Scaling human-centered machine translation. Preprint, arXiv:2207.04672. is all you need! In Proceedings of the 27th International Conference on Computational Linguistics, pages 831–844, Santa Fe, New Mexico, USA. Association for Computational Linguistics. [↗](https://arxiv.org/abs/2207.04672)
7. Deniz Gorur, Antonio Rago, and Francesca Toni. 2024. Can large language models perform relation-based argument mining? Preprint, arXiv:2402.11243. [↗](https://arxiv.org/abs/2402.11243)
8. Mohammad M. Khader, AbdulGabbar Al-Sharafi, Mohamad Hamza Al-Sioufy, Wajdi Zaghouani, and Ali Al-Zawqari. 2024. [Munazarat 1.0: A corpus of arabic competitive debates](https://md.research.elamur.ai/free/en/papers/khader-2024-munazarat-1-0-corpus-arabic/). In Proceedings of the 6th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT) with Shared Tasks on Arabic LLMs Hallucination and Dialect to MSA Machine Translation @ LREC-COLING 2024, pages 20–30, Torino, Italia. ELRA and ICCL.
9. Mohamed Bayan Kmainasi, Rakif Khan, Ali Ezzat Shahroor, Boushra Bendou, Maram Hasanain, and Firoj Alam. 2024. Native vs non-native language prompting: A comparative analysis. Preprint, arXiv:2409.07054. [↗](https://arxiv.org/abs/2409.07054)
10. Hao Li, Viktor Schlegel, Yizheng Sun, Riza Batista-Navarro, and Goran Nenadic. 2025. Large language models in argument mining: A survey. Preprint, arXiv:2506.16383. [↗](https://arxiv.org/abs/2506.16383)
11. Christian Stab and Iryna Gurevych. 2017. [Parsing argumentation structures in persuasive essays](https://md.research.elamur.ai/free/en/papers/stab-2017-parsing-argumentation-structures-persuasive-essays/). Computational Linguistics, 43(3):619–659.
12. Orith Toledo-Ronen, Matan Orbach, Yonatan Bilu, Artem Spector, and Noam Slonim. 2020. Multilingual argument mining: Datasets and analysis. Preprint, arXiv:2010.06432. [↗](https://arxiv.org/abs/2010.06432)
13. Anar Yeginbergen, Maite Oronoz, and Rodrigo Agerri. 2024. Argument mining in data scarce settings: Cross-lingual transfer and few-shot techniques. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 11687–11699, Bangkok, Thailand. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Anar+Yeginbergen%2C+Maite+Oronoz%2C+and+Rodrigo+Agerri%2E+2024%2E+Argument+mining+in+data+scarce+settings%3A+Cross%2Dlingual+transfer+and+few%2Dshot+techniques%2E+In+Proceedings+of+the+62nd+Annual+Meeting+of+the+Asso)
14. Yuning Ding, Julian Lohmann, Nils-Jonathan Schaller, Thorben Jansen, and Andrea Horbach. 2024. Transfer learning of argument mining in student essays. In Proceedings of the 19th Workshop on Innovative Use of NLP for Building Educational Applications (BEA 2024), pages 439–449, Mexico City, Mexico. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Yuning+Ding%2C+Julian+Lohmann%2C+Nils%2DJonathan+Schaller%2C+Thorben+Jansen%2C+and+Andrea+Horbach%2E+2024%2E+Transfer+learning+of+argument+mining+in+student+essays%2E+In+Proceedings+of+the+19th+Workshop+on+Innovative)
15. 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 Technologies, pages 644–648, Atlanta, Georgia. Association for Computational Linguistics. [find](https://scholar.google.com/scholar?q=Chris+Dyer%2C+Victor+Chahuneau%2C+and+Noah+A%2E+Smith%2E+2013%2E+A+simple%2C+fast%2C+and+effective+reparameterization+of+IBM+model+2%2E+In+Proceedings+of+the+2013+Conference+of+the+North+American+Chapter+of+the+Assoc)
16. Steffen Eger, Johannes Daxenberger, Christian Stab, and Iryna Gurevych. 2018. Cross-lingual argumentation mining: Machine translation (and a bit of projection) [find](https://scholar.google.com/scholar?q=Steffen+Eger%2C+Johannes+Daxenberger%2C+Christian+Stab%2C+and+Iryna+Gurevych%2E+2018%2E+Cross%2Dlingual+argumentation+mining%3A+Machine+translation+%28and+a+bit+of+projection%29)

*Updated 2 November 2026*