AI pipeline transcribes medieval legal texts with 88% accuracy

Researchers have developed an open-source pipeline to transcribe medieval English legal manuscripts, which are written in a highly abbreviated form of medieval Latin. The system uses neural networks for segmentation and handwriting recognition, achieving 79% word accuracy on a dataset of 4029 lines. Further improvements were made using an n-gram language model and by having Gemini Pro correct errors, boosting accuracy to 88%. The pipeline has been integrated into a web portal to make these historical legal documents more accessible. AI

影响 This research could enable broader access to historical legal texts, potentially uncovering new insights and aiding legal scholarship.

排序理由 This is a research paper detailing a new method for transcribing historical documents using AI. [lever_c_demoted from research: ic=1 ai=1.0]

在 arXiv cs.CV 阅读 →

• arXiv
• TrOCR
• Gemini Pro

• 论文
• 产品

AI 生成摘要 · Google Gemini · 来自 1 个来源。 我们如何撰写摘要 →