---
configs:
  - config_name: default
    data_files:
      - split: train
        path: swedish_newspaper_segments_hf.parquet
dataset_info:
  features:
    - name: image
      dtype: image
    - name: text
      dtype: string
    - name: page_filename
      dtype: string
    - name: segment_number
      dtype: int32
---

# Swedish Newspapers 1818-1870

Processed segment-level OCR dataset built from **Swedish newspapers 1818-1870** with page segmentation from PDF and ground-truth text from manual transcription.

## Data Citation Information

Språkbanken Text (2020). *Swedish newspapers 1818-1870* (updated: 2020-05-26). [Data set]. Språkbanken Text. https://doi.org/10.23695/9bnq-xc71

## Dataset Structure

This parquet file contains the following columns:

- `image` (`image`): cropped segment image bytes
- `text` (`string`): transcription text for the segment
- `page_filename` (`string`): source page ID (example: `bib11924799_18190814_145432_64_0002`)
- `segment_number` (`int32`): segment number within the page

## Basic Statistics


- Rows: **4,811**
- Distinct pages: **100**
- Parquet size: (~1.37 GB)