OCR Collection
Each reading edition is one web page: the page image beside the text, linked column by column. Two AI vision models (Claude Opus and Codex) read every page on their own, and the places where they disagree are shaded, so you can see where the machine is least sure. Each edition also has search, readings above the characters, a built-in proofreading mode and a short guided tour (the ? button).
These are machine transcriptions, not proofread. They show what AI agents can and cannot do with difficult material. The editions are made with my multi-agent vision OCR workflow.
增廣幼學須知鰲頭雜字大全 (1678)
- url
- https://www.kwokleongtang.net/projects/ocr/zengguang-youxue-xuzhi.html
- note
- A Fujian household encyclopedia and picture glossary with a very complex layout: several texts run side by side in upper and lower registers. The edition follows each text across pages and shows picture-glossary cards, tables, diagrams and map labels, with pinyin and Jyutping. About 47,500 characters in 30 texts.
- source
- 文萃堂, Fujian, 1678?; Harvard-Yenching Library (Curiosity, Chinese Rare Books)
- status
- Machine transcription by two readers; differences shaded, not settled; not proofread
- tags
- woodblock, Chinese, encyclopedia, picture glossary
安南風俗冊 (An Nam phong tục sách)
- url
- https://www.kwokleongtang.net/projects/ocr/an-nam-phong-tuc-sach.html
- note
- A Hán-script manuscript on Vietnamese customs by 段展 Đoàn Triển (Mai Viên), written in semi-cursive hand. The edition gives Hán-Việt readings above the characters.
- source
- Manuscript stamped VIETNAMIEN A 45, from the photo plates in the printed edition (pp. 114–193; plates 12a–12b missing)
- status
- Machine transcription by two readers; differences shaded, not settled; not proofread
- tags
- manuscript, Vietnamese, Hán-Việt