Here is a clean Hugging Face dataset description text you can use:
Dataset Card
Dataset Summary
This dataset is built from a large-scale digitization effort using documents sourced from the National Archives. Over 2,500 PDF files were collected, containing approximately 65,000 pages in total. Since the original documents were scanned and varied in quality, the text extraction required a robust OCR pipeline.
Google Cloud Document AI was used to process every page. Each PDF was converted into page-level text using Document AI’s high-accuracy OCR models, preserving layout structure where possible. The resulting dataset consists of text files paired with their original filenames, providing a clean and uniform representation of the archival materials.
Data Structure
Each entry contains:
- file_name: The original PDF or derived text file name.
- text: The full OCR-extracted text from that file.
Intended Use
This dataset can support research in historical document analysis, language modeling on archival material, OCR benchmarking, information retrieval, and downstream NLP tasks involving long-form or noisy text.
Additional Notes
All OCR output is provided as-is from Google Cloud Document AI, and the dataset reflects the quality and structure of the original scanned documents.