← Back to Life in London

How this Digital Version was Produced

This digital edition of Pierce Egan’s Life in London (1821) was produced between March and July 2026. It was made mostly by artificial intelligence: AI models read the scanned pages, transcribed and marked up the text, checked the transcription, and wrote the software that turned it into the web page, EPUB and ODT versions. One person, Tom Gally, started and directed the project, made the editorial decisions, and checked a selection of the doubtful readings by eye. This page explains who—and what—did each part of the work.

Who did what

Please note: the text has not been read through in full by a human proofreader. It has been checked repeatedly by AI against the page images, and samples of it have been compared word by word with the scans, but some errors may remain. The scans are the authority wherever the two differ.

Sources

The text was transcribed from the 1869–70 reprint of Life in London published by John Camden Hotten, which reproduces Egan’s 1821 text and also prints its original page numbers. Three scans were used:

• a black-and-white scan of the Hotten reprint from Cornell University Library, made available through the Internet Archive — the main source for the text;
• a colour scan of a copy held by Brigham Young University, used for comparison;
• a separate colour scan of the plates, from which the illustrations were taken.

Only Egan’s own book is included: the dedication, contents, list of illustrations, the fifteen chapters with their footnotes, the coloured plates and the engraved music of “Corinthian Tom’s Song.” Hotten’s 1869 introduction and his publisher’s advertisements were left out.

How it was done, step by step

10–11 March 2026 — the first version

1. Planning. Tom Gally set out the goals: a faithful text keeping the book’s small capitals, italics, footnotes and verse; the colour illustrations with their captions as text; and Egan’s original page numbering. Claude mapped the structure of the scans and wrote the processing scripts.
2. Choosing an AI model to read the pages. Five AI models were tested on three difficult sample pages: Google’s Gemini 3.1 Pro and Gemini 2.5 Pro, Anthropic’s Claude Sonnet 4.6 and Claude Opus 4.6, and OpenAI’s GPT-5.4. Gemini 3.1 Pro was best at small capitals and nested footnotes, so it was chosen to transcribe the book; Claude Sonnet 4.6, from a different company, was chosen to check that work.
3. Illustrations. The colour plates by I. R. and G. Cruikshank were extracted from the scans at high resolution and matched with their captions.
4. Transcription. Each of the 417 text pages was sent as an image to Gemini 3.1 Pro, which returned the text with its formatting marked up. This took about five hours. Blank pages and black-and-white copies of the plates were skipped.
5. AI proofreading. Claude Sonnet 4.6 compared each transcribed page with its image and listed suspected errors: 669 in all, mostly small capitals and italics but also 133 wrong words. The 565 it was most confident about were applied automatically; the rest were set aside for checking.
6. Assembly and output. The pages were combined into a single scholarly file in TEI-XML, the standard format for digital editions of texts. Running heads were separated from the text, footnotes joined across pages, curly quotation marks added, and words hyphenated across pages rejoined. From this file the scripts generate the web page, the EPUB e-book and the ODT word-processor file.

21–28 March 2026 — publication and review

7. The web edition was published on GitHub Pages.
8. Claude analysed the proofreading suggestions that had not yet been applied and applied the clear-cut ones. For 32 doubtful cases it built review pages showing each passage beside a crop of the scan, and Tom Gally checked these by eye; his decisions were then applied.
9. Claude explored transcribing the 17 pages of engraved sheet music into modern notation. Neither the AI models nor music-recognition software could read the old engravings reliably enough.

16–17 July 2026 — completion

10. Claude, working through the remaining tasks largely on its own, compared the last 32 doubtful readings with the scans and decided them (many turned out to be ink specks misread as punctuation).
11. It found and restored text the transcription had silently dropped from the contents and illustrations lists, and checked every suspect paragraph break at a page boundary against the scans (13 lost breaks restored, 20 false ones removed).
12. Eight randomly chosen pages were compared word by word with the scans; no misread words were found, and 13 small formatting differences were corrected.
13. The 17 pages of sheet music were added as facsimile images, each followed by the words of the song.
14. The TEI-XML file was made to validate against the TEI P5 standard, an automatic regression check was added to the scripts, and all three formats were regenerated.

Editorial decisions

• Egan’s spelling, punctuation, slang and cant are kept as printed; nothing has been modernized and there is no glossary.
• Printer’s quirks are kept where the scans show them, such as the space Egan’s printer put before ! and ?. Spaces before colons and semicolons have been removed.
• The page numbers in the contents and illustrations lists are those of the 1821 edition.
• Footnotes appear at the end of each chapter in the web version.

Source files

Everything used to make this edition—the transcribed pages, the TEI-XML file, the scripts, and the project’s working notes—is in the public repository github.com/tkgally/lifelondon.

This page was drafted by Claude (Anthropic) in November 2026 from the project’s records.

← Back to Life in London