Elamur Research⌕
☰
home / papers / 2026 / Overview of EvaHan2026: The First International Evaluation of Ancient Chinese OCR and Layout Analysis

Overview of EvaHan2026: The First International Evaluation of Ancient Chinese OCR and Layout Analysis

Paper · 2026 · Dongbo Wang, Dongmei Zhu, Jieqiong Li et al.

Summary

EvaHan2026 is the fifth edition of the EvaHan shared-task series on ancient Chinese, held with LT4HALA 2026 @ LREC 2026. Earlier editions covered segmentation, translation, punctuation and NER. This is the first edition to target the visual side of digitization: turning scanned pages into text and structure. The organizers defined three separate tasks:

  • Task A: recognition of block-engraved text.
  • Task B: layout analysis of complex pages.
  • Task C: transcription of handwritten manuscripts.

The data were manually annotated under unified rules for character variants, which the authors identify as the main source of errors. Thirteen teams registered, and four completed every task in the closed track. Top systems reached about 97% accuracy on engraved text and about 96% on handwriting. Layout detection lagged well behind, with a best mAP near 59%.

Comparing participants, the authors find that end-to-end multimodal models and layout-then-OCR pipelines each have distinct weaknesses. They recommend a staged hybrid design that combines layout detection, reading-order reconstruction and region-level OCR. The benchmark provides a reproducible basis for comparing OCR systems on historical Chinese documents.

Key points

  • Data sources: Datasets A and B come from the Siku Quanshu. Dataset C draws on varied manuscripts, including Qing diplomatic archives, private poetry collections, calligraphic copies, the Thirteen Classics and Buddhist scriptures.
  • Data size: each task has 5,000 training images and 200 test images. In Dataset B, about 27.46% of pages interleave images and text.
  • Metrics: Tasks A and C are scored mainly by CER, with F1 and NED as supporting measures. Task B is scored mainly by mAP, with IoU and F1; it covers four classes (text blocks, illustrations, book margins, seals).
  • Best results: TJU scored 0.9736 on Task A and 0.9571 on Task C, using a Qwen2.5-VL-based approach plus HistLayout-DETR layout pre-segmentation. WHU-SAI led Task B with an mAP of 0.5941, and the best reported IoU was 76.38%.
  • Baseline: a zero-shot Qwen2.5-VL baseline scored an mAP of 0.00 on Task B. In-domain fine-tuning reduced, but did not remove, two failure modes: hallucinated output and over-normalization of character variants.

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

Also on the byline: Beijing Foreign Studies University · Nanjing Agricultural University.

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksComputer Vision and Pattern Recognition · 2,869 worksEngineering · 133 worksMedia Technology · 34 works

Cite

Cite as Dongbo Wang, Dongmei Zhu, Jieqiong Li, Chang Liu, Ruifeng Wu, Fan Yang, Zhu Yue, Xin Gao, Zhao Zhixiao, Xue Zhao, Zhongzheng Wu, Liu Liu and Li Bin. 2026. Overview of EvaHan2026: The First International Evaluation of Ancient Chinese OCR and Layout Analysis. Proceedings of the Fourth Workshop on Language Technologies for Historical and Ancient Languages (LT4HALA 2026) @ LREC 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/3a35qx3n2fg9

@inproceedings{wang2026overview,
  title = {{Overview of EvaHan2026: The First International Evaluation of Ancient Chinese OCR and Layout Analysis}},
  author = {Dongbo Wang, Dongmei Zhu, Jieqiong Li, Chang Liu, Ruifeng Wu, Fan Yang, Zhu Yue, Xin Gao, Zhao Zhixiao, Xue Zhao, Zhongzheng Wu, Liu Liu and Li Bin},
  booktitle = {Proceedings of the Fourth Workshop on Language Technologies for Historical and Ancient Languages (LT4HALA 2026) @ LREC 2026},
  year = {2026},
  doi = {10.63317/3a35qx3n2fg9},
  url = {https://doi.org/10.63317/3a35qx3n2fg9}
}

References

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

  1. Li, B., Y. Yuan, J. Lu, M. Feng, C. Xu, W. Qu, and D. Wang. 2022. The First International Ancient Chinese Word Segmentation and POS Tagging Bakeoff: Overview of the EvaHan 2022 Evaluation Campaign. Proceedings of the Second Workshop on Language Technologies for Historical and Ancient Languages, page… find
  2. Wang, D., L. Lin, Z. Zhao, W. Ye, K. Meng, W. Sun, L. Zhao, X. Zhao, S. Shen, W. Zhang, and B. Li. 2023. EvaHan2023: Overview of the First International Ancient Chinese Translation Bakeoff. Proceedings of ALT2023: First Workshop on Ancient Language Translation, pages 1–14. find
  3. Li, B., B. Chang, Z. Xu, M. Feng, C. Xu, W. Qu, S. Shen, and D. Wang. 2024. Overview of EvaHan2024: The First International Evaluation on Ancient Chinese Sentence Segmentation and Punctuation. Proceedings of the Third Workshop on Language Technologies for Historical and Ancient Languages (LT4HALA 20… find
  4. Li, B., B. Chang, R. Liu, X. Zhao, S. Shen, L. Liu, Y. Zhu, Z. Xu, W. Qu, and D. Wang. 2025. Overview of EvaHan2025: The First International Evaluation on Ancient Chinese Named Entity Recognition. Proceedings of the Second Ancient Language Processing Workshop associated with NAACL 2025, pages 97–105… find
  5. Bai, J. Z., S. Bai, S. S. Yang, S. J. Wang, S. N. Tan, P. Wang, J. Y. Lin, C. Zhou, and J. R. Zhou. 2023. Qwen-VL: A versatile vision-language model for understanding, localization, text reading, and beyond. arXiv preprint arXiv:2308.12966.
  6. Carion, N., F. Massa, G. Synnaeve, N. Usunier, A. Kirillov, and S. Zagoruyko. 2020. End-to-end object detection with transformers. In Proceedings of the European Conference on Computer Vision (ECCV), pages 213–229, Cham.
  7. Springer. Hu, E. J., Y. L. Shen, P. Wallis, Z. Y. Allen-Zhu, Y. Z. Li, S. A. Wang, L. Wang, and W. Z. Chen. 2021. LoRA: Low-rank adaptation of large language models. arXiv preprint arXiv:2106.09685.
  8. Kim, G., T. Hong, M. Yim, J. Nam, J. Park, J. Yim, W. Hwang, S. Yun, D. Han, and S. Park. 2021. OCR-free document understanding transformer. In Proceedings of the European Conference on Computer Vision (ECCV).
  9. LeCun, Y., L. Bottou, Y. Bengio, and P. Haffner. 1998. Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11):2278–2324. find
  10. Li, B., B. L. Chang, R. L. Liu, X. Zhao, S. Shen, L. H. Liu, Y. Zhu, Z. X. Xu, W. G. Qu, and D. B. Wang. 2025. Overview of EvaHan2025: The first international evaluation on ancient Chinese named entity recognition. In Proceedings of the Second Workshop on Ancient Language Processing, pages 156–164,… find
  11. Li, B., B. L. Chang, Z. X. Xu, M. X. Feng, C. Xu, W. G. Qu, S. Shen, and D. B. Wang. 2024. Overview of EvaHan2024: The first international evaluation on ancient Chinese sentence segmentation and punctuation. In Proceedings of the Third Workshop on Language Technologies for Historical and Ancient Lan… find
  12. ELRA and ICCL. Li, B., Y. G. Yuan, J. Y. Lu, M. X. Feng, C. Xu, W. G. Qu, and D. B. Wang. 2022. The first international ancient Chinese word segmentation and POS tagging bakeoff: Overview of the EvaHan 2022 evaluation campaign. In Proceedings of the Second Workshop on Language Technologies for Histo… find
  13. European Language Resources Association. Lin, T. Y., M. Maire, S. Belongie, L. Bourdev, R. Girshick, J. Hays, P. Perona, D. Ramanan, C. L. Zitnick, and P. Dollár. 2014. Microsoft COCO: Common objects in context. In Proceedings of the European Conference on Computer Vision (ECCV), pages 740–755, Cham…
  14. Springer. Liu, H., C. Li, Q. Wu, and Y. J. Lee. 2024. Visual instruction tuning. In Advances in Neural Information Processing Systems (NeurIPS), 36.
  15. Long, S. B., S. Y. Qin, D. Panteleev, A. Bissacco, Y. Fujii, and M. Raptis. 2022. Towards end-to-end unified scene text detection and layout analysis. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 1039–1049. find
  16. Lu, H., W. Liu, B. Zhang, B. Wang, K. Dong, B. Liu, J. X. Sun, T. Z. Ren, Z. S. Li, H. Yang, Y. F. Sun, C. Q. Deng, H. W. Xu, Z. D. Xie, and C. Ruan. 2024. DeepSeek-VL: Towards real-world vision-language understanding. arXiv preprint arXiv:2403.05525.
  17. Rafailov, R., A. Sharma, E. Mitchell, C. D. Manning, S. Ermon, and C. Finn. 2024. Direct preference optimization: Your language model is secretly a reward model. In Advances in Neural Information Processing Systems (NeurIPS), 36.
  18. Reul, C., D. Christ, A. Hartelt, N. Balbach, M. Wehner, U. Springmann, C. Wick, C. Grundig, A. Büttner, and F. Puppe. 2019. OCR4all - an open-source tool providing a (semi-)automatic OCR workflow for historical printings. Applied Sciences, 9(22):4853. Shao, Z. H., P. Y. Wang, Q. H. Zhu, R. X. Xu, J.… find
  19. Song, X. Bi, H. W. Zhang, M. C. Zhang, Y. K. Li, Y. Wu, and D. Y. Guo. 2024. DeepSeekMath: Pushing the limits of mathematical reasoning in open language models. arXiv preprint arXiv:2402.03300.
  20. Shen, Z., R. Zhang, M. Dell, B. C. G. Lee, J. Carlson, and W. Li. 2021. LayoutParser: A unified toolkit for deep learning based document image analysis. In Proceedings of the International Conference on Document Analysis and Recognition (ICDAR), pages 131– 146. find
  21. Springer International Publishing. Shi, B., X. Bai, and C. Yao. 2015. An end-to-end trainable neural network for image-based sequence recognition and its application to scene text recognition. IEEE Transactions on Pattern Analysis and Machine Intelligence, 39:2298–2304. find
  22. Shi, Y., C. Liu, D. Peng, C. Jian, J. Huang, and L. Jin. 2023. M5HisDoc: A large-scale multi-style Chinese historical document analysis benchmark. In Advances in Neural Information Processing Systems (NeurIPS), 36. find
  23. Wang, D. B., L. T. Lin, Z. X. Zhao, W. H. Ye, K. Meng, W. L. Sun, L. Z. Zhao, X. Zhao, S. Shen, W. Zhang, and B. Li. 2023. EvaHan2023: Overview of the first international ancient Chinese translation bakeoff. In Proceedings of the Ancient Language Translation Workshop (ALT2023), pages 1–14, Macao SAR… find
  24. Wang, T., Y. Wu, R. Rubino, and C. Baumgartner. 2019. Radical aggregation network for few-shot offline handwritten Chinese character recognition. Pattern Recognition Letters, 125:821–827. find
  25. Wang, X., Y. Wang, B. Li, and D. Wang. 2024. Xunzi: A Chinese ancient text large language model. In Proceedings of the Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024), pages 11984–11993. find
  26. Xu, Y., F. Yin, Z. X. Zhang, and C. L. Liu. 2018. Multi-task layout analysis for historical handwritten documents using fully convolutional networks. In Proceedings of the 27th International Joint Conference on Artificial Intelligence (IJCAI’18), pages 1057–1063. find
  27. AAAI Press. Yu, H. Y., Y. C. Wu, F. Shi, L. Liao, J. H. Lu, X. D. Ge, H. Wang, M. H. Zhuo, X. C. Wu, X. Fei, H. Feng, G. Z. Tang, A. L. Wang, H. S. Zhu, Y. F. He, Q. H. Liang, L. Y. Meng, C. Feng, C. Huang, J. Q. Tang, and B. Li. 2025. Benchmarking vision-language models on Chinese ancient documents…

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/wang-2026-overview-evahan2026-first-international-evaluation/