wjbmattingly/medieval-letters-htr-synthetic

Medieval Letters Synthetic Dataset

A clean synthetic dataset of 8317 medieval manuscript pages with transcriptions.

Dataset Description

This dataset contains high-quality synthetic medieval manuscript pages featuring:

• Synthetic manuscript images (375x500px) with realistic textures and ink variations
• Abbreviated transcriptions as they appear on the manuscripts (e.g., "omn;" for "omnibus")
• Original transcriptions before abbreviation rules were applied
• Page XML containing the full ALTO XML with layout and text information

Features

• image: The synthetic manuscript page image (PNG, 375x500px)
• transcription: Text as it appears on the manuscript with medieval abbreviations
• full_page_transcription: Original text before abbreviations were applied
• page_xml: Raw ALTO XML containing layout and transcription data

Usage

from datasets import load_dataset

# Load the dataset
dataset = load_dataset("path/to/medieval-letters-synthetic")

# Access a sample
sample = dataset['train'][0]

# Display image and transcriptions side by side
print("Image size:", sample['image'].size)
print("Abbreviated:", sample['transcription'][:100], "...")
print("Original:", sample['full_page_transcription'][:100], "...")

# Show the image
sample['image'].show()

# Access XML data
xml_content = sample['page_xml']
print("XML length:", len(xml_content), "characters")

Image and Text Side-by-Side

This dataset is perfect for training models that need to correlate visual manuscript content with textual transcriptions:

import matplotlib.pyplot as plt

sample = dataset['train'][0]

fig, (ax1, ax2) = plt.subplots(1, 2, figsize=(15, 6))

# Show image
ax1.imshow(sample['image'])
ax1.set_title('Manuscript Image')
ax1.axis('off')

# Show transcriptions
text_content = f"Abbreviated:\n{sample['transcription'][:200]}...\n\nOriginal:\n{sample['full_page_transcription'][:200]}..."
ax2.text(0.05, 0.95, text_content, transform=ax2.transAxes, fontsize=10, 
         verticalalignment='top', wrap=True)
ax2.set_title('Transcriptions')
ax2.axis('off')

plt.tight_layout()
plt.show()

Applications

Perfect for:

• Handwritten Text Recognition (HTR) model training
• OCR development for historical documents
• Medieval paleography research
• Abbreviation expansion studies
• Image-to-text alignment tasks

Technical Details

• Image format: PNG, 375x500px resolution, embedded as bytes
• Text encoding: UTF-8
• Language: Medieval Latin
• Script: Gothic/medieval style
• Abbreviations: Includes ligatures, contractions, and Tironian notes
• XML format: ALTO XML with layout and transcription information

License

CC-BY-4.0

Citation

@dataset{medieval_letters_synthetic,
  title={Medieval Letters Synthetic Dataset},
  year={2024},
  description={Clean synthetic medieval manuscript dataset with 8317 pages},
  license={CC-BY-4.0}
}