digital-history-bielefeld/image-text_anglicana-legal-texts

Dataset Card for FLOW Project: Medieval Anglicana Legal Texts (13th-14th Century)

Dataset Summary

This dataset contains 21,579 lines of high-quality Ground Truth data for Handwritten Text Recognition (HTR). It was created by the FLOW Project at Bielefeld University (Digital History) to facilitate the training of HTR models for medieval administrative and legal documents.

The dataset consists of image snippets (lines) paired with their corresponding transcriptions and the line coordinates. The material focuses on legal records from the 13th and 14th centuries written in Anglicana script and Latin language.

Dataset Structure

Each entry in the dataset includes:

• image: The image snippet of a single line.
• text: The diplomatic transcription of the line.
• line_coords & line_baseline: Geometric metadata for precise localization.
• region_type: Information about the document structure (e.g., marginalia vs. main text).
• filename & project_name: Provenance information for tracking the source document.

Source Data & Provenance

The digital images were originally provided by the AALT Society (Archive of Available Leases and Titles). The original physical records are held by The National Archives (TNA), Kew, UK.

Included Documents

This dataset was compiled from a variety of legal documents. For a detailed list of all TNA shelfmarks and foliants included in this dataset, please refer to the supplementary file: 👉 (https://huggingface.co/datasets/digital-history-bielefeld/image-text_anglicana-legal-texts/blob/main/documents.csv)

Licensing and Usage

This dataset is published under the following terms:

1. Transcriptions: Published under the Open Government Licence v3.0.
2. Image Snippets: The National Archives (TNA) has granted special permission to publish these image snippets alongside the transcription work to facilitate HTR research and training.
3. Usage: This data is intended for non-commercial academic research and the advancement of Open Science only.

Attribution Statement:

"The National Archives (TNA) recognizes the work of transcription which has been done for the project. These image snippets are reproduced by permission of The National Archives, UK, to facilitate handwritten text recognition research. Original records held at The National Archives, Kew."

Transcription Guidelines

The transcriptions in this dataset follow a diplomatic approach [oder "semi-diplomatic", je nachdem was zutrifft].

Key features of our transcription policy:

• Abbreviations: Expanded.
• Character normalization: Original spelling was preserved (e.g., u and v were kept as found whilst only i was used and no j).
• Capitalization: Follows the manuscript's original usage.
• Special signs: Paragraph marks were transcribed as ⁋, the middle dot ‧ was used as well as the regular dot . and the rare comma. No continuation marks were transcribed except when the text used them, then they were noted as -. The ∧ symbol was used to mark spots for interlinear additions and corrections when the sign could be found in the text, which was not always the case.

Related Models

A model trained on this dataset (utilizing the TrOCR architecture) can be found here: dh-unibe/trocr-essoins-middle-latin

Citation

If you use this dataset in your research, please cite it as follows:

@dataset{wilde_kuhlmann_flow_2026,
  author       = {Melvin Wilde and Christopher Kuhlmann},
  title        = {FLOW Project: Medieval Anglicana HTR Training Data (13th-14th Century)},
  year         = {2026},
  publisher    = {Hugging Face},
  url          = {https://huggingface.co/datasets/digital-history-bielefeld/image-text_anglicana-legal-texts}
}