histde/1k-books-corpus

Public Domain German Books (1800–1879)

A page-level dataset of 1,111 German books (299,610 pages) printed between 1800 and 1879, digitised by the Münchener DigitalisierungsZentrum (MDZ) of the Bayerische Staatsbibliothek. Every page comes with its full-resolution scan, the hOCR output, plain text extracted from that hOCR, and the IIIF Image API description of the scan.

All images, info.jsons and hOCR were downloaded from the official APIs.

Dataset Structure

One row = one page. Book-level metadata is repeated on every page of the book, so rows are self-contained and can be filtered or grouped without joins.

column	type	description
id	string	page identifier, <mdz_id>_<page> (e.g. bsb00116182_00001)
mdz_id	string	Object identifier of the book (e.g. bsb00116182)
page	int32	1-based page (canvas) number within the book
title_display	string	title of the book
pers_inst_facet	string	persons / institutions (authors etc.), comma-separated
date_sort	int32	publication year
number_scans	int32	number of pages of the book
licence	string	rights statement of the scans (always the CC Public Domain Mark)
object_details	string	Catalogue page of the book
iiif_manifest	string	IIIF Presentation API v2 manifest of the book
image_url	string	IIIF Image API URL of the full-resolution scan
ocr_url	string	URL of the hOCR file for the page
text	string	plain text of the page, extracted from the hOCR (see below)
hocr	string	the complete hOCR document of the page (words with bounding boxes and confidences)
width, height	int32	pixel size of the scan, from info.json
info	string	the IIIF Image API info.json of the scan
image	Image	the scan (JPEG, full resolution, unmodified)

text is derived from the hOCR structure: words of an ocr_line are joined with spaces, lines of a paragraph (ocr_par / ocrx_block) with newlines, and paragraphs are separated by a blank line. Tokenisation and hyphenation are kept as produced by the OCR (e.g. punctuation is a separate token: Zeiten , so). Blank pages have an empty text.

Dataset Stats

decade	books	pages
1800–1809	139	48,831
1810–1819	139	43,377
1820–1829	139	36,620
1830–1839	139	42,103
1840–1849	139	28,411
1850–1859	139	32,599
1860–1869	139	42,759
1870–1879	138	24,910
total	1,111	299,610

Licensing

The digitised works are in the public domain and the scans are published by the Bavarian State Library under the CC Public Domain Mark 1.0.