--- title: "Democratizing the medieval English legal tradition — Zhang 2026" description: "The earliest centuries of the common law survive as millions of handwritten pages in heavily abbreviated medieval… 30% WER. 6 references." authors: ["Michael Zhang", "Elise Wang", "Charlotte Whatley", "Seth Strickland", "Dylan Bannon"] year: 2026 venue: "arXiv:2605.00977" source: "https://arxiv.org/abs/2605.00977" doi: "10.48550/arxiv.2605.00977" arxiv: "2605.00977" licence: "CC BY-SA 4.0 (arXiv)" licence_url: "http://creativecommons.org/licenses/by-sa/4.0/" summary_licence: "Elamur Cosmic License" canonical: "https://research.elamur.ai/free/en/papers/zhang-2026-democratizing-medieval-english-legal-tradition/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/zhang-2026-democratizing-medieval-english-legal-tradition/) [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/27/) / Democratizing the medieval English legal tradition # Democratizing the medieval English legal tradition **Authors:** [Michael Zhang](https://md.research.elamur.ai/free/en/authors/michael-zhang/), Elise Wang, Charlotte Whatley, Seth Strickland and Dylan Bannon · **Year:** 2026 · **Venue:** [arXiv:2605.00977](https://md.research.elamur.ai/free/en/venues/lecture-notes-in-computer-science/) ([Lecture notes in computer science 2026](https://md.research.elamur.ai/free/en/venues/lecture-notes-in-computer-science/2026/)) · **Publisher:** [Springer Science+Business Media](https://md.research.elamur.ai/free/en/publishers/springer-science-business-media/) · **Licence:** [CC BY-SA 4.0 (arXiv)](http://creativecommons.org/licenses/by-sa/4.0/) **Links:** [Publisher's page](https://arxiv.org/abs/2605.00977) · [PDF](https://arxiv.org/pdf/2605.00977) · [DOI 10.48550/arxiv.2605.00977](https://doi.org/10.48550/arxiv.2605.00977) · [arXiv 2605.00977](https://arxiv.org/abs/2605.00977) · [OpenAlex](https://openalex.org/W7160237171) **Citations:** 0 citations (OpenAlex, 2026-10-14) ## Summary The earliest centuries of the common law survive as millions of handwritten pages in heavily abbreviated medieval Latin. Only a few dozen specialists can read them, so the material is digitized but effectively closed. The authors built a training set of 4029 transcribed lines from 193 criminal and civil cases. The cases come from rolls on the AALT archive and date mainly from 1272–1461. The set is released in PageXML. On this data they trained an open pipeline with two stages: - R-Blla handles line segmentation. - A CNN+LSTM recognizer decoded with CTC produces text with abbreviations already expanded. The base system reaches about 79% word accuracy. Adding an n-gram language model to decoding raises this to 82%, and having Gemini Pro 3 correct the output raises it to 88%. However, the LLM correction can conceal errors a reader would otherwise notice. TrOCR matches the word accuracy but has worse character accuracy. Used alone, frontier LLMs mostly fail at transcription. The pipeline runs as a public web portal, glyphmachina.com, so non-specialists can work with the legal archive. ## Key points - **Dataset:** 4029 lines (3602 train, 427 test) from 193 cases, written in cursiva anglicana / chancery cursive. Sources are King's Bench (KB27), Common Pleas (CP40) and eyre (JUST1) rolls, plus 51 AALT palaeography example lines. At most about three cases were taken from any single roll. - **Accuracy:** word accuracy is 79% for CNN+LSTM+CTC, 82% with the n-gram (bigram) decoder and 88% after Gemini Pro 3 post-correction. - **TrOCR vs the main model:** TrOCR guesses grammatical Latin words that are wrong for the case. Its word accuracy is similar, but its character accuracy is worse. - **Gemini Pro 3 transcribing on its own:** - Only 3 test cases came out under 30% WER, and none beat the trained model. - 2 cases returned nothing. - 9 cases had the wrong number of lines because of hallucinated text. - GPT 5.1, Claude Sonnet 4.5 and Grok 4.1 refused or hallucinated. - **Generalization:** - The model performs well on JUST2/JUST3 rolls and on unseen KB/CP rolls. - It is weaker on Exchequer (E) rolls, which contain many numbers and different vocabulary. - It does better on documents older than the training period than on later ones. - It does not handle vernacular-language documents. ## Interesting ideas - An LLM corrector raises headline accuracy but can turn visible errors into fluent invisible ones, so the authors advise keeping the raw model output beside the corrected version. - Plausible-but-wrong guesses (TrOCR) are scored as more harmful than transparent partial errors, because they block a human from recovering the true reading; character-level accuracy captures this better than word accuracy alone. - A small domain n-gram model inside the decoder resolves visually ambiguous sequences (e.g., runs of identical minim strokes) using local context, giving a cheap and inspectable gain. - The sample was built at the case level and spread widely: few items per source, deliberately varied by handwriting and case type, so a small set stands in for a huge stable corpus. - A simple structural check, comparing the expected and returned number of lines, exposed LLM hallucination without reading the content. ## For Mova How these ideas could serve Mova, our small open language model — its files, stages and tests: - **I1** (learn): when an LLM teacher tier repairs a parse or frame, store the witness parsers' original output and the diff, so a reviewer can see what changed and where the confident fix was unsupported. - **I2** (tokens): for unknown or garbled words, the lemmatizer and tagger should emit a visibly uncertain or partial analysis rather than snapping to a common real lemma, and evaluation should penalize such confident substitutions. - **I3** (tokens): domain-specific n-gram statistics could break ties in POS or morphology for ambiguous forms in formulaic texts (tests, word problems) while staying explainable. - **I4** (tree): when picking sentences for silver treebank data or manual checking, cap items per source document and stratify by genre or author, to cover the dialect space with fewer annotations. - **I5** (back): in the text → graph → text compressor, compare sentence and entity counts between input and regenerated text as a quick hallucination or omission alarm before deeper semantic comparison. *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 - [Carnegie Mellon University](https://md.research.elamur.ai/free/en/universities/carnegie-mellon-university/) — University, Pittsburgh, United States · 539 works · 179,533 citations - [University of Chicago](https://md.research.elamur.ai/free/en/universities/university-of-chicago/) — University, Chicago, United States · 45 works · 50,617 citations - [University of Wisconsin–Madison](https://md.research.elamur.ai/free/en/universities/university-of-wisconsin-madison/) — University, Madison, United States · 42 works · 48,486 citations Also on the byline: [California State University, Fullerton](https://ror.org/02avqqw26). 📚 Shelves (6): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [👁️ Computer Vision and Pattern Recognition](https://md.research.elamur.ai/free/en/shelves/computer-science/computer-vision-and-pattern-recognition/) · 2,869 works · [🤝 Social Sciences](https://md.research.elamur.ai/free/en/shelves/social-sciences/) · 1,772 works · [📚 Literature and Literary Theory](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/literature-and-literary-theory/) · 519 works · [🤝 Political Science and International Relations](https://md.research.elamur.ai/free/en/shelves/social-sciences/political-science-and-international-relations/) · 182 works 🧭 Topics (2): [English](https://md.research.elamur.ai/free/en/topics/english/) · 2,263 works · [⚖️ Legal AI](https://md.research.elamur.ai/free/en/topics/legal-ai/) · 134 works ## Cite Cite as: Michael Zhang, Elise Wang, Charlotte Whatley, Seth Strickland and Dylan Bannon. 2026. Democratizing the medieval English legal tradition. arXiv:2605.00977. The paper itself, not this page. BibTeX from the library's record of the work. ```bibtex @misc{zhang2026democratizing, title = {{Democratizing the medieval English legal tradition}}, author = {Michael Zhang, Elise Wang, Charlotte Whatley, Seth Strickland and Dylan Bannon}, year = {2026}, doi = {10.48550/arxiv.2605.00977}, eprint = {2605.00977}, archivePrefix = {arXiv}, url = {https://doi.org/10.48550/arxiv.2605.00977} } ``` - [← previous DeID-Clinic: A Risk-Aware Pseudonymization Framework for Clinical Text De-identification and Re-identification Risk Assessment](https://md.research.elamur.ai/free/en/papers/paul-2026-deid-clinic-risk-aware-pseudonymization/) - [next → Demystifying Funding: Reconstructing a Unified Dataset of the UK Funding Lifecycle](https://md.research.elamur.ai/free/en/papers/thorne-2026-demystifying-funding-reconstructing-unified-dataset/) - [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/27/) ## References 6 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. 1. Aguilar, S.T.: TRIDIS: A comprehensive medieval and early modern corpus for HTR and NER. CoRR abs/2503.22714 (2025). https://doi.org/10.48550/ ARXIV.2503.22714, https://doi.org/10.48550/arXiv.2503.22714 2. Aguilar, S.T., Jolivet, V.: Handwritten text recognition for documentary medieval manuscripts. Journal of Data Mining & Digital Humanities Historical Documents and automatic text recognition, 6 (Dec 2023). https://doi.org/10. 46298/jdmdh.10484, https://jdmdh.episciences.org/10484 3. Clérice, T., Pinche, A., Vlachou-Efstathiou, M., Chagué, A., Camps, J.B., Levenson, M.G., Brisville-Fertin, O., Boschetti, F., Fischer, F., Gervers, M., Boutreux, A., Manton, A., Gabay, S., O’Connor, P., Haverals, W., Kestemont, M., Vandyck, C., Kiessling, B.: Catmus medieval: A multilingual large-scale cross-century dataset in latin script for handwritten text recognition and beyond. In: Barney Smith, E.H., Liwicki, M., Peng, L. (eds.) Document Analysis and Recognition - ICDAR 2024. pp. 174–194. [↗](https://doi.org/10.48550/arxiv.2503.22714) 2. Springer Nature Switzerland, Cham (2024) 4. Coll Ardanuy, M., Berganzo-Besga, I., Sarobe, R., Cuadrada, C.: Evaluating handwritten text recognition in medieval notarial manuscripts: A new dataset and comprehensive analysis. In: Yin, X.C., Karatzas, D., Lopresti, D. (eds.) Document Analysis and Recognition – ICDAR 2025. pp. 340–357. [find](https://scholar.google.com/scholar?q=Springer+Nature+Switzerland%2C+Cham+%282024%29+4%2E+Coll+Ardanuy%2C+M%2E%2C+Berganzo%2DBesga%2C+I%2E%2C+Sarobe%2C+R%2E%2C+Cuadrada%2C+C%2E%3A+Evaluating+handwritten+text+recognition+in+medieval+notarial+manuscripts%3A+A+new+dataset+and+) 3. Springer Nature Switzerland, Cham (2026) 5. Diaz, D.H., Qin, S., Ingle, R.R., Fujii, Y., Bissacco, A.: Rethinking text line recognition models. CoRR abs/2104.07787 (2021), https://arxiv.org/abs/2104.07787 6. Fujitake, M.: Dtrocr: Decoder-only transformer for optical character recognition. In: IEEE/CVF Winter Conference on Applications of Computer Vision, WACV 2024, Waikoloa, HI, USA, January 3-8, 2024. pp. 8010–8020. [↗](https://arxiv.org/abs/2104.07787) 4. IEEE (2024). https://doi.org/10.1109/WACV57701.2024.00784, https://doi.org/ 10.1109/WACV57701.2024.00784 Democratizing the medieval English legal tradition 7. Garrido-Munoz, C., Rios-Vila, A., Calvo-Zaragoza, J.: Handwritten Text Recognition: A Survey. arXiv e-prints arXiv:2502.08417 (Feb 2025). https://doi.org/10. 48550/arXiv.2502.08417 8. Kahle, P., Colutto, S., Hackl, G., Mühlberger, G.: Transkribus - a service platform for transcription, recognition and retrieval of historical documents. In: 2017 14th IAPR International Conference on Document Analysis and Recognition (ICDAR). vol. 04, pp. 19–24 (2017). https://doi.org/10.1109/ICDAR.2017.307 9. Kang, L., Riba, P., Rusiñol, M., Fornés, A., Villegas, M.: Pay attention to what you read: Non-recurrent handwritten text-line recognition. Pattern Recognit. 129, 108766 (2022). https://doi.org/10.1016/J.PATCOG.2022.108766, https://doi.org/ 10.1016/j.patcog.2022.108766 10. Kiessling, B.: A modular region and text line layout analysis system. In: 2020 17th International Conference on Frontiers in Handwriting Recognition (ICFHR). pp. 313–318 (2020). https://doi.org/10.1109/ICFHR2020.2020.00064 11. Kim, G., Govindaraju, V.: A lexicon driven approach to handwritten word recognition for real-time applications. IEEE Trans. Pattern Anal. Mach. Intell. 19(4), 366–379 (Apr 1997). https://doi.org/10.1109/34.588017, https://doi.org/10.1109/ 34.588017 12. Li, M., Lv, T., Chen, J., Cui, L., Lu, Y., Florêncio, D.A.F., Zhang, C., Li, Z., Wei, F.: Trocr: Transformer-based optical character recognition with pre-trained models. In: Williams, B., Chen, Y., Neville, J. (eds.) Thirty-Seventh AAAI Conference on Artificial Intelligence, AAAI 2023, Thirty-Fifth Conference on Innovative Applications of Artificial Intelligence, IAAI 2023, Thirteenth Symposium on Educational Advances in Artificial Intelligence, EAAI 2023, Washington, DC, USA, February 7-14, 2023. pp. 13094–13102. [↗](https://doi.org/10.1109/wacv57701.2024.00784) 5. AAAI Press (2023). https://doi.org/10.1609/AAAI. V37I11.26538, https://doi.org/10.1609/aaai.v37i11.26538 13. Nockels, J., Gooding, P., Ames, S., Terras, M.: Understanding the application of handwritten text recognition technology in heritage contexts: a systematic review of transkribus in published research. Archival Science 22(3), 367– 392 (2022). https://doi.org/10.1007/s10502-022-09397-0, https://doi.org/10.1007/ s10502-022-09397-0 14. Pugh, R.B.: The duration of criminal trials in medieval england. In: Manchester, A., Ives, E.W. (eds.) Law, Litigants and the Legal Profession. pp. 104–115. [↗](https://doi.org/10.1609/aaai) 6. Royal Historical Society (1983) 15. Puigcerver, J., Mocholí, C.: Pylaia. https://github.com/jpuigcerver/PyLaia (2018) 16. Shi, B., Bai, X., Yao, C.: An end-to-end trainable neural network for image-based sequence recognition and its application to scene text recognition. IEEE Trans. Pattern Anal. Mach. Intell. 39(11), 2298–2304 (2017). https://doi.org/10.1109/ TPAMI.2016.2646371, https://doi.org/10.1109/TPAMI.2016.2646371 17. Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A.N., Kaiser, L., Polosukhin, I.: [Attention is all you need](https://md.research.elamur.ai/free/en/papers/vaswani-2017-attention-all-need/). In: Guyon, I., von Luxburg, U., Bengio, S., Wallach, H.M., Fergus, R., Vishwanathan, S.V.N., Garnett, R. (eds.) Advances in Neural Information Processing Systems 30: Annual Conference on Neural Information Processing Systems 2017, December 4-9, 2017, Long Beach, CA, USA. pp. 5998–6008 (2017), https://proceedings.neurips.cc/paper/ 2017/hash/3f5ee243547dee91fbd053c1c4a845aa-Abstract.html [↗](https://doi.org/10.1109/tpami.2016.2646371) *Updated 30 October 2026*