humument-data

Per-page OCR data for the 1892 one-volume Chapman & Hall edition of W. H. Mallock's A Human Document — the novel Tom Phillips treated to make A Humument. This package is the data layer consumed by humument; most users should use that library instead of fetching these files directly.

Layout

db/
  catalog.json          { pages: [...], chapters: [...] }
  search-index.json     { token: [[pageNum, count], ...] }
  pages/pNNNN.json.gz   one per printed page (1–367), gzipped JSON:
                        { meta, words[], gutters[], docks[], graph[] }

Page files are gzip-compressed (the raw set is ~159MB; gzipped ~26MB so CDNs will serve it). Decode in the browser with DecompressionStream:

const res = await fetch('https://cdn.jsdelivr.net/npm/[email protected]/db/pages/p0040.json.gz');
const page = await new Response(res.body.pipeThrough(new DecompressionStream('gzip'))).json();

words[] carry text, pixel bboxes (on the 1400×2100 normalized page images — see humument-images), line index, OCR confidence, punctuation prefix/suffix, and NLP annotations (POS, lemma, frequency, rarity). gutters/docks/graph describe the inter-word whitespace and a navigable graph for drawing Phillips-style "rivers".

Provenance

OCR'd (Apple Vision) from the Internet Archive scan ahumandocumenta04mallgoog. The 1892 novel is in the public domain; this packaging and the derived data are MIT-licensed. pNNNN is the printed book page = the A Humument page.