Elamur Research⌕
☰
home / papers / 2026 / GATE-Reranker: A Strong Arabic Cross-Encoder for Document Reranking

GATE-Reranker: A Strong Arabic Cross-Encoder for Document Reranking

Paper · 2026 · Omer Nacar, Omar Elshehy, Mohamed Zaytoon et al.

Summary

Arabic search and retrieval-augmented generation (RAG) systems often use a second-stage neural reranker. Most strong rerankers are multilingual or English-trained, and their score reliability on Arabic has received little study. The authors present GATE-Reranker, a compact Arabic cross-encoder. It starts from the GATE Arabic embedding model and is fine-tuned on large-scale Arabic mMARCO-style triplets. Each query and passage are encoded together with full self-attention, and a linear head on the [CLS] representation outputs a relevance score. Training uses pointwise regression with MSE loss on positive (1) and negative (0) pairs, chosen because it tends to give calibrated scores. The model was tested on three Arabic benchmarks: binary relevance, a fixed five-candidate reranking task, and the large mMARCO dev set. Its ranking quality is close to strong multilingual rerankers, and its scores are much better calibrated and more clearly separated. It also runs with low memory and latency on a Tesla T4 GPU. This allows fixed relevance thresholds for filtering documents in RAG pipelines, which can reduce noise passed to the generator.

Key points

  • Evaluation data:
    • Arabic mMARCO dev: 55,560 queries over about 8.8M passages.
    • Arabic-Reranking-Triplet-5-Eval: 500 queries, each with 1 relevant and 4 non-relevant passages.
    • Ar-Reranking-Eval: 500 binary-labelled samples.
  • Calibration versus MiniLM-L12-v2: ECE was 0.0316 vs 0.4995, and Brier score was 0.0077 vs 0.4994.
  • Score separation: the mean positive–negative margin was 0.9382 for GATE-Reranker versus 0.0002 for MiniLM. FPR@95TPR was 0.0 versus 0.6453.
  • Variants: GATE-Reranker-V1 is the efficiency-oriented model. A larger scaling variant was also analysed to map the accuracy–efficiency trade-off.
  • Limitations:
    • The context window is 512 tokens.
    • End-to-end recall was not evaluated.
    • Some benchmarks are near ceiling, so they discriminate poorly between models.
    • Biases may be inherited from mMARCO, including limited dialect coverage.

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

🎓Saarland UniversitySaarbrücken · Germany148 works · 18,287 citations🎓Alexandria UniversityAlexandria · Egypt39 works · 287 citations🎓Arab International UniversityGhabāghib · Syria13 works · 46 citations🎓Tuwaiq AcademyRiyadh · Saudi Arabia11 works · 7 citations

 Fields, topics and concepts

📚 Shelves 6The 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 worksComputer Vision and Pattern Recognition · 2,869 worksInformation Systems · 1,437 works

🧭 Topics 2Research subjects and languages the work is about.Semitic languages · 2,908 worksInformation retrieval & search · 473 works

Cite

Cite as Omer Nacar, Omar Elshehy, Mohamed Zaytoon and Khloud Al Jallad. 2026. GATE-Reranker: A Strong Arabic Cross-Encoder for Document Reranking. The 7th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT7) with 5 Shared Tasks.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/2y297wwcf77y

@inproceedings{nacar2026gate,
  title = {{GATE-Reranker: A Strong Arabic Cross-Encoder for Document Reranking}},
  author = {Omer Nacar, Omar Elshehy, Mohamed Zaytoon and Khloud {Al Jallad}},
  booktitle = {The 7th Workshop on Open-Source Arabic Corpora and Processing Tools (OSACT7) with 5 Shared Tasks},
  year = {2026},
  doi = {10.63317/2y297wwcf77y},
  url = {https://doi.org/10.63317/2y297wwcf77y}
}

References

18 references, 10 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Payal Bajaj, Daniel Campos, Nick Craswell, Li Deng, Jianfeng Gao, Xiaodong Liu, Rangan Majumder, Andrew McNamara, Bhaskar Mitra, Tri Nguyen, et al. 2016. Ms marco: A human generated machine reading comprehension dataset. arXiv preprint arXiv:1611.09268. ↗
  2. Chris Burges, Tal Shaked, Erin Renshaw, Ari Lazier, Matt Deeds, Nicole Hamilton, and Greg Hullender. 2005. Learning to rank using gradient descent. In Proceedings of the 22nd International Conference on Machine Learning, ICML ’05, page 89–96, New York, NY, USA. Association for Computing Machinery.
  3. Jianlv Chen, Shitao Xiao, Peitian Zhang, Kun Luo, Defu Lian, and Zheng Liu. 2024. Bge m3-embedding: Multi-lingual, multi-functionality, multi-granularity text embeddings through self-knowledge distillation. find
  4. Hervé Déjean, Stéphane Clinchant, and Thibault Formal. 2024. A thorough comparison of cross-encoders and llms for reranking splade. arXiv preprint arXiv:2403.10407. ↗
  5. Yanhua Huang, Yuqi Chen, Xiong Cao, Rui Yang, Mingliang Qi, Yinghao Zhu, Qingchang Han, Yaowei Liu, Zhaoyu Liu, Xuefeng Yao, Yuting Jia, Leilei Ma, Yinqi Zhang, Taoyu Zhu, Liujie Zhang, Lei Chen, Weihang Chen, Min Zhu, Ruiwen Xu, and Lei Zhang. 2025. Towards large-scale generative ranking. ↗
  6. Vladimir Karpukhin, Barlas Oguz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih. 2020. Dense passage retrieval for open-domain question answering. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 6769–6781, Online. Association for Computational Linguistics. ↗
  7. Omar Khattab and Matei Zaharia. 2020. Colbert: Efficient and effective passage search via contextualized late interaction over bert. In Proceedings of the 43rd International ACM SIGIR conference on research and development in Information Retrieval, pages 39–48. ↗
  8. Chaofan Li, Zheng Liu, Shitao Xiao, and Yingxia Shao. 2023. Making large language models a better foundation for dense retrieval. find
  9. Hamza Shahid Mohaned A. Rashad. 2025. The arabic rag leaderboard. urlhttps://huggingface.co/spaces/Navid-AI/The-Arabic-Rag-Leaderboard. ↗
  10. Omer Nacar, Anis Koubaa, Serry Sibaee, Yasser Al-Habashi, Adel Ammar, and Wadii Boulila. 2025. Gate: General arabic text embedding for enhanced semantic textual similarity with matryoshka representation learning and hybrid loss training. arXiv preprint arXiv:2505.24581. ↗
  11. Rodrigo Nogueira and Kyunghyun Cho. 2019. Passage re-ranking with bert. ArXiv, abs/1901.04085. ↗
  12. Rodrigo Nogueira, Wei Yang, Kyunghyun Cho, and Jimmy Lin. 2019. Multi-stage document ranking with bert. ↗
  13. Zhen Qin, Rolf Jagerman, Kai Hui, Honglei Zhuang, Junru Wu, Le Yan, Jiaming Shen, Tianqi Liu, Jialu Liu, Donald Metzler, Xuanhui Wang, and Michael Bendersky. 2024. Large language models are effective text rankers with pairwise ranking prompting. In Findings of the Association for Computational Linguistics: NAACL 2024, pages 1504–1518, Mexico City, Mexico. Association for Computational Linguistics. ↗
  14. Nandan Thakur, Nils Reimers, Andreas Rücklé, Abhishek Srivastava, and Iryna Gurevych. 2021. Beir: A heterogenous benchmark for zero-shot evaluation of information retrieval models. arXiv preprint arXiv:2104.08663. ↗
  15. Feng Wang, Yuqing Li, and Han Xiao. 2025. jina-reranker-v3: Last but not late interaction for listwise document reranking. ↗
  16. Yining Wang, Liwei Wang, Yuanzhi Li, Di He, and Tie-Yan Liu. 2013. A theoretical analysis of ndcg type ranking measures. In Conference on learning theory, pages 25–54. PMLR. ↗
  17. Lee Xiong, Chenyan Xiong, Ye Li, Kwok-Fung Tang, Jialin Liu, Paul N. Bennett, Junaid Ahmed, and Arnold Overwijk. 2021. Approximate nearest neighbor negative contrastive learning for dense text retrieval. In International Conference on Learning Representations. ↗
  18. Yanzhao Zhang, M Li, D Long, X Zhang, H Lin, B Yang, P Xie, A Yang, D Liu, J Lin, et al. 2025. Qwen3 embedding: Advancing text embedding and reranking through foundation models, 2025. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/nacar-2026-gate-reranker-strong-arabic-cross/