Skip to content

Dataset · storytracer

German historical newspaper archive (13B words)

storytracer/German-PD-Newspapers

Massive corpus of OCR texts from public domain old German newspapers for training language models.

What's inside

Contains 13 billion words (1 to 10 million records) in German, obtained via OCR from historical newspapers. Data is sourced from the Deutsche Digitale Bibliothek repository and distributed under free CC0 and Public Domain licenses.

What you can do with it

  • Pre-train or fine-tune German-language LLMs from scratch
  • Adapt language models to historical German spelling and vocabulary
  • Train OCR post-processing systems for archival documents
Tasks
Text generation
Languages
German
Size
1M–10M rows
Downloads this month
820
Likes
8

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("storytracer/German-PD-Newspapers")