Elamur Research⌕
☰
home / papers / 2026 / Enabling Stroke-Level Structural Analysis of Hieroglyphic Scripts without Language-Specific Priors

Enabling Stroke-Level Structural Analysis of Hieroglyphic Scripts without Language-Specific Priors

Paper · 2026 · Fuwen Luo, Zihao Wan, Ziyue Wang et al.

Summary

Logographic and hieroglyphic scripts carry meaning in how strokes are arranged inside each character. Current LLMs treat characters as opaque tokens, and MLLMs see them only as pixel grids, so neither captures stroke structure. Earlier structural methods are usually tied to one script and depend on costly annotation or predefined stroke inventories. The authors present HieroSA (Hieroglyphic Stroke Analyzer), a reinforcement-learning framework that teaches an MLLM to turn a binarized character image into a set of line segments in normalized coordinates. Training needs no stroke labels or linguistic rules. The reward is computed from the bitmap alone and favors segments that stay on black pixels and cover as much of the glyph as possible. Experiments span modern logographic and ancient scripts. They show that the inferred segments capture internal structure and semantics, generalize across writing systems, and help downstream glyph recognition (OCR) and structure-based character exploration. This makes HieroSA a candidate tool for computational graphematics and decipherment work, including scripts with no agreed stroke system, such as Oracle Bone Script.

Key points

  • Each stroke is two endpoints in a normalized coordinate space. Curves are approximated by combining several segments. Decompositions may differ from conventional human stroke definitions.
  • The reward has three stages:
    • Validity check: points are sampled along each segment, and the segment counts as invalid if any point falls off the black pixels.
    • Coverage estimate: rays are cast along the normal and tangent directions, with abnormally long extensions truncated by a λ-based threshold.
    • Aggregation: the per-stroke results are combined into the final reward.
  • In an ablation on the Chinese training set (100 steps), using 3 or 4 points per stroke performed worse than 2 points on reward, coverage, and invalid-stroke rate. The two-point design was therefore kept.
  • The paper discusses Oracle Bone Script and Dongba among the target scripts, alongside modern systems such as Chinese characters. Code is released at github.com/THUNLP-MT/HieroSA.
  • The authors list four limitations:
    • sensitivity to noisy or structurally complex training data;
    • only moderate-scale models and datasets;
    • a single-model setup;
    • some manual inspection still needed to pick useful exploration results.

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

🎓Tsinghua UniversityBeijing · China333 works · 87,019 citations

 Fields, topics and concepts

📚 Shelves 5The 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 works

Cite

Cite as Fuwen Luo, Zihao Wan, Ziyue Wang, Yaluo Liu, Pau Tong Lin Xu, Xuanjia Qiao, Xiaolong Wang, Peng Li and Yang Liu. 2026. Enabling Stroke-Level Structural Analysis of Hieroglyphic Scripts without Language-Specific Priors. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.1383

@inproceedings{luo2026enabling,
  title = {{Enabling Stroke-Level Structural Analysis of Hieroglyphic Scripts without Language-Specific Priors}},
  author = {Fuwen Luo, Zihao Wan, Ziyue Wang, Yaluo Liu, Pau Tong Lin Xu, Xuanjia Qiao, Xiaolong Wang, Peng Li and Yang Liu},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.1383},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.1383}
}

References

55 references, 8 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Emre Aksan, Thomas Deselaers, Andrea Tagliasacchi, and Otmar Hilliges. 2020. Cose: Compositional stroke embeddings. Advances in Neural Information Processing Systems, 33:10041–10052. find
  2. James P Allen. 2000. Middle Egyptian: An introduction to the language and culture of hieroglyphs. Cambridge University Press. find
  3. T George Allen. 1951. Egyptian grammar, being an introduction to the study of hieroglyphs. find
  4. Yannis Assael, Thea Sommerschield, Brendan Shillingford, Mahyar Bordbar, John Pavlopoulos, Marita Chatzipanagiotou, Ion Androutsopoulos, Jonathan Prag, and Nando De Freitas. 2022. Restoring and attributing ancient texts using deep neural networks. Nature, 603(7900):280–283. find
  5. Shuai Bai, Yuxuan Cai, Ruizhe Chen, Keqin Chen, Xionghui Chen, Zesen Cheng, Lianghao Deng, Wei Ding, Chang Gao, Chunjiang Ge, and 1 others. 2025. Qwen3-vl technical report. Preprint, arXiv:2511.21631.
  6. Xiaojun Bi, Shuo Li, Junyao Xing, Ziyue Wang, Fuwen Luo, Weizheng Qiao, Lu Han, Ziwei Sun, Peng Li, and Yang Liu. 2025. DongbaMIE: A Multimodal Information Extraction Dataset for Evaluating Semantic Understanding of Dongba Pictograms. In Findings of the Association for Computational Linguistics: EMN… find
  7. William G Boltz. 1986. Early chinese writing. World Archaeology, 17(3):420–436. find
  8. Gerlof Bouma. 2009. Normalized (pointwise) mutual information in collocation extraction. Proceedings of GSCL, 30:31–40. find
  9. Shaosheng Cao, Wei Lu, Jun Zhou, and Xiaolong Li. 2018. cw2vec: Learning chinese word embeddings with stroke n-gram information. In Proceedings of the AAAI conference on artificial intelligence, volume 32. find
  10. Yingfa Chen, Chenlong Hu, Cong Feng, Chenyang Song, Shi Yu, Xu Han, Zhiyuan Liu, and Maosong Sun. 2025. Multi-modal multi-granularity tokenizer for chu bamboo slips. In Proceedings of the 31st International Conference on Computational Linguistics, pages 6201–6211, Abu Dhabi, UAE. Association for Com… find
  11. Zongze Chen, Wenxia Yang, and Xin Li. 2023. Stroke-based autoencoders: Self-supervised learners for efficient zero-shot chinese character recognition. Applied Sciences, 13(3):1750. find
  12. W Vivian Davies. 1990. Egyptian hieroglyphs. JT Hooker et al., Reading the Past, pages 75–135. find
  13. Alexey Dosovitskiy. 2020. An image is worth 16x16 words: Transformers for image recognition at scale. arXiv preprint arXiv:2010.11929.
  14. Jiageng Fan. 2014. A Study of Characters in Chinese and Japanese, including Semantic Shift. Ph.D. thesis, University of Canterbury. find
  15. Hao Fei, Shengqiong Wu, Hanwang Zhang, Tat-Seng Chua, and Shuicheng Yan. 2024. Vitron: A unified pixel-level vision llm for understanding, generating, segmenting, editing. Advances in neural information processing systems, 37:57207–57239. find
  16. Haisu Guan, Huanxin Yang, Xinyu Wang, Shengwei Han, Yongge Liu, Lianwen Jin, Xiang Bai, and Yuliang Liu. 2024. Deciphering oracle bone language with diffusion models. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 15554– 1556… find
  17. David Ha and Douglas Eck. 2017. A neural representation of sketch drawings. arXiv preprint arXiv:1704.03477. — [arXiv:1704.03477](https://arxiv.org/abs/1704.03477) find
  18. James W Heisig. 2008. Remembering the Kanji: A systematic guide to reading the japanese characters, volume 2. University of Hawaii Press. find
  19. John S Henderson. 1997. The World of the Ancient Maya. Cornell University Press. find
  20. Yiheng Huang, Shuang She, Zewei Wei, Jianmin Lin, Ming Yang, and Wenyin Liu. 2025. StrokeNet: Unveiling How to Learn Fine-Grained Interactions in Online Handwritten Stroke Classification. In International Conference on Document Analysis and Recognition, pages 200–217. Springer. find
  21. Qi Jia, Xin Fan, Meiyu Yu, Yuqing Liu, Dingrong Wang, and Longin Jan Latecki. 2020. Coupling Deep Textural and Shape Features for Sketch Recognition. In Proceedings of the 28th ACM International Conference on Multimedia, pages 421–429. find
  22. Guangyuan Jiang, Matthias Hofer, Jiayuan Mao, Lionel Wong, Joshua B Tenenbaum, and Roger P Levy. 2024. Finding Structure in Logographic Writing with Library Learning. arXiv preprint arXiv:2405.06906. — [arXiv:2405.06906](https://arxiv.org/abs/2405.06906) find
  23. Terry Joyce. 2011. The significance of the morphographic principle for the classification of writing systems. Written Language & Literacy, 14(1):58–81. find
  24. Ilker Kesen, Jonas F Lotz, Ingo Ziegler, Phillip Rust, and Desmond Elliott. 2025. Multilingual Pretraining for Pixel Language Models. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 29582–29599. find
  25. Tomohiko Morioka. 2015. Multiple-policy Character Annotation Based on CHISE. Journal of the Japanese Association for Digital Humanities, 1(1):86–106. find
  26. Ricardo Muñoz Sánchez. 2024. When Hieroglyphs Meet Technology: A Linguistic Journey through Ancient Egypt Using Natural Language Processing. In Proceedings of the Third Workshop on Language Technologies for Historical and Ancient Languages (LT4HALA) @ LREC-COLING-2024, pages 156– 169, Torino, Italia… find
  27. Minh Nguyen, Gia H. Ngo, and Nancy F. Chen. 2020. Hierarchical character embeddings: Learning phonological and semantic representations in languages of logographic origin using recursive neural networks. IEEE/ACM Transactions on Audio, Speech, and Language Processing, 28:461–473. find
  28. Rémi Pautrat, Daniel Barath, Viktor Larsson, Martin R Oswald, and Marc Pollefeys. 2023. DeepLSD: Line Segment Detection and Refinement with Deep Image Gradients. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 17327–17336. find
  29. Hengnian Qi, Hao Yang, Zhaojiang Wang, Jiabin Ye, Qiuyi Xin, Chu Zhang, and Qing Lang. 2025. Ancient-GlyphNet: an advanced deep learning framework for detecting ancient Chinese characters in complex scene. Artificial Intelligence Review, 58(3):88. find
  30. Taku Kudo and John Richardson. 2018. SentencePiece: A simple and language independent subword tokenizer and detokenizer for neural text processing. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, pages 66–71.
  31. Alec Radford, Jong Wook Kim, Chris Hallacy, Aditya Ramesh, Gabriel Goh, Sandhini Agarwal, Girish Sastry, Amanda Askell, Pamela Mishkin, Jack Clark, and 1 others. 2021. Learning Transferable Visual Models from Natural Language Supervision. In International conference on machine learning, pages 8748–8…
  32. Sander Land and Max Bartolo. 2024. Fishing for Magikarp: Automatically Detecting Under-trained Tokens in Large Language Models. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 11631–11646.
  33. Phillip Rust, Jonas F Lotz, Emanuele Bugliarello, Elizabeth Salesky, Miryam de Lhoneux, and Desmond Elliott. 2023. Language Modelling with Pixels. In The Eleventh International Conference on Learning Representations. find
  34. Dayiheng Liu, Kexin Yang, Qian Qu, and Jiancheng Lv. 2019. Ancient–Modern Chinese Translation with a New Large Training Dataset. ACM Transactions on Asian and Low-Resource Language Information Processing (TALLIP), 19(1):1–13. find
  35. Mengting Liu, Guoying Liu, Yongge Liu, and Qingju Jiao. 2020. Oracle bone inscriptions recognition based on deep convolutional neural network. Journal of image and graphics, 8(4):114–119. find
  36. Xiaojuan Liu, Xu Han, Shanxiong Chen, Weijia Dai, and Qiuyue Ruan. 2024. Ancient Yi Script Handwriting Sample Repository. Scientific Data, 11(1):1183. find
  37. Qin Lu. 2019. Computers and Chinese Writing Systems. In The Routledge handbook of Chinese applied linguistics, pages 461–482. Routledge. find
  38. Rico Sennrich, Barry Haddow, and Alexandra Birch. 2016. Neural machine translation of rare words with subword units. In Proceedings of the 54th annual meeting of the association for computational linguistics (volume 1: long papers), pages 1715–1725.
  39. Zhihong Shao, Peiyi Wang, Qihao Zhu, Runxin Xu, Junxiao Song, Xiao Bi, Haowei Zhang, Mingchuan Zhang, Y. K. Li, Y. Wu, and Daya Guo. 2024. DeepSeekMath: Pushing the Limits of Mathematical Reasoning in Open Language Models. arXiv preprint arXiv:2402.03300.
  40. Johannes Singer, Katja Seeliger, and Martin N Hebart. 2020. The Representation of Object Drawings and Sketches in Deep Convolutional Neural Networks. In NeurIPS 2020 Workshop SVRHM. find
  41. Thea Sommerschield, Yannis Assael, John Pavlopoulos, Vanessa Stefanak, Andrew Senior, Chris Dyer, John Bodel, Jonathan Prag, Ion Androutsopoulos, and Nando De Freitas. 2023. Machine learning for ancient languages: A survey. Computational Linguistics, 49(3):703–747.
  42. Iago Suárez, José M Buenaposada, and Luis Baumela. 2022. ELSED: Enhanced Line Segment Drawing. Pattern Recognition, 127:108619. find
  43. Michael Tschannen, Alexey Gritsenko, Xiao Wang, Muhammad Ferjad Naeem, Ibrahim Alabdulmohsin, Nikhil Parthasarathy, Talfan Evans, Lucas Beyer, Ye Xia, Basil Mustafa, and 1 others. 2025. SigLIP 2: Multilingual Vision-Language Encoders with Improved Semantic Understanding, Localization, and Dense Feat… find
  44. Pengjie Wang, Kaile Zhang, Xinyu Wang, Shengwei Han, Yongge Liu, Jinpeng Wan, Haisu Guan, Zhebin Kuang, Lianwen Jin, Xiang Bai, and 1 others. 2024. An Open Dataset for Oracle Oone Script Recognition and Decipherment. arXiv preprint arXiv:2401.15365. — [arXiv:2401.15365](https://arxiv.org/abs/2401.15365) find
  45. Qian Yu, Yongxin Yang, Feng Liu, Yi-Zhe Song, Tao Xiang, and Timothy M Hospedales. 2017. Sketch-a-Net: A Deep Neural Network that Beats Humans. International journal of computer vision, 122(3):411– 425. find
  46. Hong-Mei Zhang and Yuan-Ming Mao. 2017. An Overview of Ancient Chinese Character Inscriptions of National Minorities. Journal of Chinese Writing Systems, 1(1):19–28. find
  47. Shouhui Zhao. 2005. Chinese Character Modernisation in the Digital Era: A Historical Perspective. Current Issues in Language Planning, 6(3):315–378. find
  48. Shu Zhou, Xin Wang, Jingwen Qiu, Wenru Bu, and Hao Wang. 2025. OracleNet: Enhancing Oracle Bone Script Recognition with Adaptive Deformation and Texture-Structure Decoupling. npj Heritage Science, 13(1):273. find
  49. Zhijun Wang, Xuebo Liu, and Min Zhang. 2022. Breaking the representation bottleneck of Chinese characters: Neural machine translation with stroke sequence modeling. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 6473–6484, Abu Dhabi, United Arab Emir… find
  50. Adelina Wei-Kwan Wong. 2018. Ancient Chinese hieroglyphs: Archetypes of transformation. Jung Journal, 12(3):54–74. find
  51. Mary B Woods and Michael Woods. 2011. Ancient Communication Technology: From Hieroglyphics to Scrolls. Twenty-First Century Books. find
  52. Xiaofeng Wu, Karl Stratos, and Wei Xu. 2025. The impact of visual information in chinese characters: Evaluating large models’ ability to recognize and utilize radicals. In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Huma… find
  53. Zongyang Xiong, Ke Qin, Haobo Yang, and Guangchun Luo. 2021. Learning chinese word representation better by cascade morphological n-gram. Neural Computing and Applications, 33(8):3757–3768. find
  54. Fuquan Yang. 2012. Living Hieroglyphics and Dongba Culture. Yunnan Education Press. find
  55. Fen Yi, Hong Liu, You Wang, Sheng Wu, Cheng Sun, Peng Feng, and Jin Zhang. 2023. Medical named entity recognition fusing part-of-speech and stroke features. Applied Sciences, 13(15):8913. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/luo-2026-enabling-stroke-level-structural-analysis/