leitro/Copiale_Lines

---
pretty_name: Copiale Lines
task_categories:
- image-to-text
language:
- de
size_categories:
- 1K<n<10K
viewer: true
---

# Copiale Lines

Copiale Lines is a line-level image-to-text dataset for historical cipher decipherment. It contains cropped line images from the Copiale manuscript paired with plaintext ground truth.

This dataset is used in the paper *Learning to Decipher from Pixels: A Case Study of Copiale* (HistoCrypt 2026).

## Dataset Structure

The dataset is split into:

- train: 1,269 samples
- valid: 175 samples
- test: 370 samples

Each split contains:

- `images/*.png`: cropped line images
- `metadata.csv`: filename and plaintext transcription

The corresponding source split files are `train.gt`, `valid.gt`, and `test.gt`, where each line is:

```text
image_id<TAB>groundtruth
```

## Example

```text
1-2.png,gesetz buchs
```

corresponds to the image:

```text
train/images/1-2.png
```

## Intended Use

This dataset is intended for research on handwritten cipher recognition, image-to-text modeling, and transcription-free decipherment.

## Citation

```bibtex
@inproceedings{kang2026learning,
title = {Learning to Decipher from Pixels: A Case Study of Copiale},
author = {Kang, Lei and De Gregorio, Giuseppe and Heil, Raphaela and Fornés, Alicia and Megyesi, Beáta},
booktitle = {International Conference on Historical Cryptology (HistoCrypt)},
year = {2026}
}
```

## Acknowledgements

This dataset is derived in part from materials related to *Decipherment of Historical Manuscripts*, a historical manuscript studied within the project ["The Copiale Cipher"](https://www.su.se/english/research/research-catalogue/research-projects/d/decipherment-of-historical-manuscripts/the-copiale-cipher) at Stockholm University. We acknowledge and thank the original project for making these resources available.

We also gratefully acknowledge financial support from **Riksbankens Jubileumsfond** under grant **M24-0028**, *"Echoes of History: Analysis and Decipherment of Historical Writings (DESCRYPT)"*, which supported the development of this dataset.