---
title: "HCMUS_TheFangs at AbjadGenEval Shared Task: Weighted"
description: "The paper describes the first-place system in the AbjadGenEval shared task. 0.93 F1. 21 references."
authors: ["Duy Minh Dao Sy", "Nguyen Chi Tran", "Trung Kiet Huynh", "Nguyen Lam Phu Quy", "Pham Phu Hoa", "Nguyen Dinh Ha Duong"]
year: 2026
venue: "Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script"
source: "https://aclanthology.org/2026.abjadnlp-1.51/"
doi: "10.18653/v1/2026.abjadnlp-1.51"
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/sy-2026-hcmus-thefangs-abjadgeneval-shared-task/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/sy-2026-hcmus-thefangs-abjadgeneval-shared-task/)

[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/page/32/) / HCMUS\_TheFangs at AbjadGenEval Shared Task: Weighted Layer Pooling with Attention Fusion for Arabic AI-Generated Text Detection

# HCMUS\_TheFangs at AbjadGenEval Shared Task: Weighted Layer Pooling with Attention Fusion for Arabic AI-Generated Text Detection

**Authors:** [Duy Minh Dao Sy](https://md.research.elamur.ai/free/en/authors/duy-minh-dao-sy/), [Nguyen Chi Tran](https://md.research.elamur.ai/free/en/authors/nguyen-chi-tran/), [Trung Kiet Huynh](https://md.research.elamur.ai/free/en/authors/trung-kiet-huynh/), [Nguyen Lam Phu Quy](https://md.research.elamur.ai/free/en/authors/nguyen-lam-phu-quy/), [Pham Phu Hoa](https://md.research.elamur.ai/free/en/authors/pham-phu-hoa/) and [Nguyen Dinh Ha Duong](https://md.research.elamur.ai/free/en/authors/nguyen-dinh-ha-duong/) · **Year:** 2026 · **Venue:** [Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script](https://md.research.elamur.ai/free/en/venues/abjadnlp/) ([Workshop on NLP for Languages Using Arabic Script 2026](https://md.research.elamur.ai/free/en/venues/abjadnlp/2026/)) · **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/2026.abjadnlp-1.51/) · [PDF](https://aclanthology.org/2026.abjadnlp-1.51.pdf) · [DOI 10.18653/v1/2026.abjadnlp-1.51](https://doi.org/10.18653/v1/2026.abjadnlp-1.51) · [OpenAlex](https://openalex.org/W7140123865)

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

## Summary
The paper describes the first-place system in the AbjadGenEval shared task. The task is binary classification of Arabic news texts as written by a human or produced by an LLM (GPT-3.5, GPT-4, Claude). The training data is heavily skewed toward machine text: 36,182 machine samples against 5,621 human ones.

The authors fine-tune mDeBERTa-v3 Base (microsoft/mdeberta-v3-base) with a two-branch pooling head:
- **Weighted Layer Pooling** learns softmax-normalized weights over the [CLS] vectors of layers 4–12.
- **Attention Pooling** aggregates token states from the final layer.

The two vectors are concatenated and passed to an MLP. Training uses 5-fold stratified cross-validation and discriminative learning rates.

The team tested more than 15 alternatives, including MixUp, adversarial training, contrastive objectives, and stylometric or discourse features. Their main finding is that the choice of encoder mattered far more than training tricks: switching from AraBERT to DeBERTa-v3 accounted for most of the gain. The final system scored 0.93 F1, 3 points ahead of the runner-up. This suggests that for detecting AI-generated text in lower-resourced languages, choosing a strong backbone should come before elaborate training schemes.

## Key points
- **Official results:** F1 0.93, accuracy 0.93, precision 0.97, recall 0.89, ranking 1st. The runner-up scored F1 0.90.
- **Backbone comparison:**
  - The AraBERT baseline reached F1 0.73, and MixUp plus adversarial training did not improve it.
  - A plain DeBERTa-v3 baseline reached 0.90, which is about a 23% relative gain.
  - The full system's 0.93 corresponds to the reported +27% relative gain over AraBERT.
- **Feature-based approaches performed poorly:** stylometric features scored F1 0.65, dediacritization 0.64, discourse-marker counting 0.63, and LLM-voting ensembles 0.62.
- **Learned layer weights rise toward upper layers:** layer 12 received 0.19, and layers 10–12 together carry about 45% of the weight. The authors read this as semantic coherence being a stronger cue than local syntax.
- **Training setup:**
  - Input is limited to 512 tokens, with diacritics and punctuation kept.
  - Learning rates are 1e-5 for the backbone (layer-wise decay 0.95) and 1e-4 for the new pooling and classifier layers.
  - Training ran for 5 epochs at batch size 8 on one H100, taking about 20 minutes per fold.
  - Out-of-fold macro-F1 was 0.98.

*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

- [Vietnam National University Ho Chi Minh City](https://md.research.elamur.ai/free/en/universities/vietnam-national-university-ho-chi-minh-city/) — University, Ho Chi Minh City, Vietnam · 21 works · 14 citations
- [Ho Chi Minh City University of Science](https://md.research.elamur.ai/free/en/universities/ho-chi-minh-city-university-of-science/) — University, Ho Chi Minh City, Vietnam · 5 works · 2 citations

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works

🧭 Topics (1): [Semitic languages](https://md.research.elamur.ai/free/en/topics/semitic-languages/) · 2,908 works

## Cite

Cite as: Duy Minh Dao Sy, Nguyen Chi Tran, Trung Kiet Huynh, Nguyen Lam Phu Quy, Pham Phu Hoa and Nguyen Dinh Ha Duong. 2026. HCMUS\_TheFangs at AbjadGenEval Shared Task: Weighted Layer Pooling with Attention Fusion for Arabic AI-Generated Text Detection. Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script.

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

```bibtex
@inproceedings{sy2026hcmus,
  title = {{HCMUS\_TheFangs at AbjadGenEval Shared Task: Weighted Layer Pooling with Attention Fusion for Arabic AI-Generated Text Detection}},
  author = {Duy Minh Dao Sy, Nguyen Chi Tran, Trung Kiet Huynh, Nguyen Lam Phu Quy, Pham Phu Hoa and Nguyen Dinh Ha Duong},
  booktitle = {Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script},
  year = {2026},
  doi = {10.18653/v1/2026.abjadnlp-1.51},
  url = {https://doi.org/10.18653/v1/2026.abjadnlp-1.51}
}
```

- [← previous HCMUS_The Fangs at AbjadStyleTransfer Shared Task: Learning to Query Style, Contrastive Representations for Zero-Shot Arabic Authorship Style Transfer](https://md.research.elamur.ai/free/en/papers/sy-2026-hcmus-fangs-abjadstyletransfer-shared-task/)
- [next → HCMUS_TheFangs at NakbaArchiveClassifier Shared Task: Foundation Models and Advanced Training Strategies for Conflict Damage Classification](https://md.research.elamur.ai/free/en/papers/sy-2026-hcmus-thefangs-nakbaarchiveclassifier-shared-task/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/32/)

## References

21 references, 15 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Shadi Abudalfa, Saad Ezzini, Ahmed Abdelali, Hamza Alami, Abdessamad Benlahbib, Salmane Chafik, Mo El-Haj, Abdelkader El Mahdaouy, Mustafa Jarrar, Salima Lamsiyah, and Hamzah Luqman. 2025. [The AraGenEval shared task on Arabic authorship style transfer and AI generated text detection](https://md.research.elamur.ai/free/en/papers/abudalfa-2025-arageneval-shared-task-arabic-authorship/). In Proceedings of The Third Arabic Natural Language Processing Conference at EMNLP 2025, Suzhou, China. Association for Computational Linguistics.
2. Ganesh Jawahar, Benoît Sagot, and Djamé Seddah. 2019. [What does bert learn about the structure of language?](https://md.research.elamur.ai/free/en/papers/jawahar-2019-what-does-bert-learn-about/) Proceedings of the 57th Annual Meeting of the ACL.
3. Thomas Lavergne, Tanguy Urvoy, and François Yvon. 2008. Detecting fake content with relative entropy scoring. Proceedings of the 2008 International Workshop on Plagiarism Analysis, Authorship Identification, and Near-Duplicate Detection. [find](https://scholar.google.com/scholar?q=Thomas+Lavergne%2C+Tanguy+Urvoy%2C+and+Fran%C3%A7ois+Yvon%2E+2008%2E+Detecting+fake+content+with+relative+entropy+scoring%2E+Proceedings+of+the+2008+International+Workshop+on+Plagiarism+Analysis%2C+Authorship+Identifi)
4. Jiucai Li and Yong Jiang. 2023. Deepfake text detection: A study of detection methods and model robustness. arXiv preprint arXiv:2303.16876. [↗](https://arxiv.org/abs/2303.16876)
5. Anonymous. 2025. Diveye: Detecting ai-generated text via fluctuating unpredictability. OpenReview Preprint. [find](https://scholar.google.com/scholar?q=Anonymous%2E+2025%2E+Diveye%3A+Detecting+ai%2Dgenerated+text+via+fluctuating+unpredictability%2E+OpenReview+Preprint%2E)
6. Wissam Antoun, Fady Baly, and Hazem Hajj. 2020. [Arabert: Transformer-based model for arabic language understanding](https://md.research.elamur.ai/free/en/papers/antoun-2020-arabert-transformer-based-model-arabic/). Proceedings of the 4th Workshop on Open-Source Arabic Corpora and Processing Tools. [↗](https://doi.org/10.48550/arxiv.2003.00104)
7. Saad Ezzini, Irfan Ahmed, Salmane Chafik, Shadi Abudalfa, Mo El-Haj, Ahmed Abdelali, Mustafa Jarrar, Nadir Durrani, Hassan Sajjad, and Farah Adeeba. 2026. [Abjadgeneval: Abjad ai generated text detection shared task for languages using arabic script at abjadnlp 2026](https://md.research.elamur.ai/free/en/papers/ezzini-2026-abjadgeneval-abjad-ai-generated-text/). In Proceedings of the 2nd Workshop on NLP for Languages Using Arabic Script (AbjadNLP 2026), co-located with the 19th Conference of the European Chapter of the Association for Computational Linguistics (EACL 2026), Rabat, Morocco.
8. Sebastian Gehrmann, Hendrik Strobelt, and Alexander M. Rush. 2019. [Gltr: Statistical detection and visualization of generated text](https://md.research.elamur.ai/free/en/papers/gehrmann-2019-gltr-statistical-detection-visualization-generated/). Proceedings of the 57th Annual Meeting of the ACL: System Demonstrations.
9. Ian J Goodfellow, Jonathon Shlens, and Christian Szegedy. 2014. [Explaining and harnessing adversarial examples](https://md.research.elamur.ai/free/en/papers/goodfellow-2014-explaining-harnessing-adversarial-examples/). arXiv preprint arXiv:1412.6572. [↗](https://arxiv.org/abs/1412.6572)
10. Nizar Habash. 2010. Introduction to arabic natural language processing. Synthesis Lectures on Human Language Technologies, 3(1):1–187. [↗](https://doi.org/10.1007/978-3-031-02139-8)
11. Pengcheng He, Jianfeng Gao, and Weizhu Chen. 2021. [Debertav3: Improving deberta using electra-style pre-training with gradient-disentangled embedding sharing](https://md.research.elamur.ai/free/en/papers/he-2021-debertav3-improving-deberta-using-electra/). arXiv preprint arXiv:2111.09543. [↗](https://arxiv.org/abs/2111.09543)
12. 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](https://md.research.elamur.ai/free/en/papers/inoue-2021-interplay-variant-size-task-type/). Proceedings of the Sixth Arabic NLP Workshop.
13. Yikang Liu, Ziyin Yao, Wenyue Li, et al. 2023. Argugpt: Evaluating, understanding and identifying argumentative essays generated by gpt models. arXiv preprint arXiv:2304.07666. [↗](https://arxiv.org/abs/2304.07666)
14. Ilya Loshchilov and Frank Hutter. 2019. Decoupled weight decay regularization. In ICLR. [find](https://scholar.google.com/scholar?q=Ilya+Loshchilov+and+Frank+Hutter%2E+2019%2E+Decoupled+weight+decay+regularization%2E+In+ICLR%2E)
15. Eric Mitchell, Yoonho Lee, Alexander Khazatsky, Christopher D. Manning, and Chelsea Finn. 2023. [Detectgpt: Zero-shot machine-generated text detection using probability curvature](https://md.research.elamur.ai/free/en/papers/mitchell-2023-detectgpt-zero-shot-machine-generated/). Proceedings of the 40th International Conference on Machine Learning.
16. Matthew E. Peters, Mark Neumann, Mohit Iyyer, Matt Gardner, Christopher Clark, Kenton Lee, and Luke Zettlemoyer. 2018. [Deep contextualized word representations](https://md.research.elamur.ai/free/en/papers/peters-2018-deep-contextualized-word-representations/). In Proceedings of the 2018 Conference of the North American Chapter of the ACL: Human Language Technologies, Volume 1 (Long Papers), pages 2227–2237.
17. Anna Rogers, Olga Kovaleva, and Anna Rumshisky. 2020. [A primer in bertology: What we know about how bert works](https://md.research.elamur.ai/free/en/papers/rogers-2020-primer-bertology-what-we-know/). Transactions of the ACL, 8:842–866. [↗](https://doi.org/10.1162/tacl_a_00349)
18. Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, Łukasz Kaiser, and Illia Polosukhin. 2017. [Attention is all you need](https://md.research.elamur.ai/free/en/papers/vaswani-2017-attention-all-need/). In Advances in Neural Information Processing Systems.
19. Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Pierric Cistac, Tim Rault, et al. 2019. Hugging face’s [transformers: State-of-the-art natural language processing](https://md.research.elamur.ai/free/en/papers/wolf-2020-transformers-state-art-natural-language/). arXiv preprint arXiv:1910.03771. [↗](https://arxiv.org/abs/1910.03771)
20. Hongyi Zhang, Moustapha Cisse, Yann N Dauphin, and David Lopez-Paz. 2017. [mixup: Beyond empirical risk minimization](https://md.research.elamur.ai/free/en/papers/zhang-2017-mixup-empirical-risk-minimization/). arXiv preprint arXiv:1710.09412. [↗](https://arxiv.org/abs/1710.09412)
21. Ganesh Jawahar, Muhammad Abdul-Mageed, and Laks V.S. Lakshmanan. 2020. [Automatic detection of machine generated text: A critical survey](https://md.research.elamur.ai/free/en/papers/jawahar-2020-automatic-detection-machine-generated-text/). Proceedings of the 28th International Conference on Computational Linguistics.

*Updated 1 November 2026*