---
license: cc-by-4.0
task_categories:
- image-to-text
language:
- fr
- en
- nl
- it
- es
- ca
pretty_name: CATMuS Medieval
size_categories:
- 100K<n<1M
tags:
- optical-character-recognition
- humanities
- handwritten-text-recognition
---
# Dataset Card for CATMuS Medieval

![Banner for the CATMuS Project](banner_catmus_medieval_centered.png)

Join our Discord to ask questions about the dataset: [![Join the Discord](https://img.shields.io/badge/CATMuS-Discord?style=flat-square&logo=discord&logoColor=%23333333&color=%235865F2)](https://discord.gg/J38xgNEsGk)

## Dataset Details

Handwritten Text Recognition (HTR) has emerged as a crucial tool for converting manuscripts images into machine-readable formats, 
enabling researchers and scholars to analyse vast collections efficiently. 
Despite significant technological progress, establishing consistent ground truth across projects for HTR tasks, 
particularly for complex and heterogeneous historical sources like medieval manuscripts in Latin scripts (8th-15th century CE), remains nonetheless challenging. 
We introduce the **Consistent Approaches to Transcribing Manuscripts (CATMuS)** dataset for medieval manuscripts, 
which offers:
1. a uniform framework for annotation practices for medieval manuscripts,
2. a benchmarking environment for evaluating automatic text recognition models across multiple dimensions thanks to rich metadata (century of production, 
language, genre, script, etc.),
3. a benchmarking environment for other tasks (such as script classification or dating approaches),
4. a benchmarking environment  and finally for exploratory work pertaining to computer vision and digital paleography around line-based tasks, such as generative approaches.

Developed through collaboration among various institutions and projects, CATMuS provides an inter-compatible dataset spanning more than 200 manuscripts and incunabula in 10 
different languages, comprising over 160,000 lines of text and 5 million characters spanning from the 8th century to the 16th.

The dataset's consistency in transcription approaches aims to mitigate challenges arising from the diversity in standards for medieval manuscript transcriptions, 
providing a comprehensive benchmark for evaluating HTR models on historical sources. 

### Dataset Description

<!-- Provide a longer summary of what this dataset is. -->

- **Curated by:** Thibault Clérice
- **Funded by:** BnF Datalab, Biblissima +, DIM PAMIR
- **Language(s) (NLP):** Middle and Old French, Middle Dutch, Catalan, Spanish, Navarese, Italian, Venitian, Old English, Latin
- **License:** CC-BY

<!--
### Dataset Sources [optional]

 Provide the basic links for the dataset.

- **Repository:** [More Information Needed]
- **Paper [optional]:** [More Information Needed]
- **Demo [optional]:** [More Information Needed]
 -->

## Uses

### Direct Use

- Handwritten Text Recognition
- Date classification
- Script classification

### Out-of-Scope Use

- Text-To-Image

## Dataset Structure

- Data contains the main `split` that is loaded through `load_dataset("CATMuS/medieval")`
- Data can be split with each manuscript inside train, val and test using the `gen_split` columns which results in a 90/5/5 split
- The image is in the `im` column, and the text in the `text` column

<!--
## Dataset Creation

### Curation Rationale

 Motivation for the creation of this dataset.

[More Information Needed]

### Source Data

 This section describes the source data (e.g. news text and headlines, social media posts, translated sentences, ...). 

#### Data Collection and Processing

<!-- This section describes the data collection and processing process such as data selection criteria, filtering and normalization methods, tools and libraries used, etc.

[More Information Needed]

#### Who are the source data producers?

<!-- This section describes the people or systems who originally created the data. It should also include self-reported demographic or identity information for the source data creators if 
this information is available. 

[More Information Needed]
 -->

### Annotations [optional]

#### Annotation process

The annotation process is described in the [dataset paper](https://inria.hal.science/hal-04453952).

#### Who are the annotators?

- Pinche, Ariane
- Clérice, Thibault
- Chagué, Alix
- Camps, Jean-Baptiste
- Vlachou-Efstathiou, Malamatenia
- Gille Levenson, Matthias
- Brisville-Fertin, Olivier
- Boschetti, Federico
- Fischer, Franz
- Gervers, Michael
- Boutreux, Agnès
- Manton, Avery
- Gabay, Simon
- Bordier, Julie
- Glaise, Anthony
- Alba, Rachele
- Rubin, Giorgia
- White, Nick
- Karaisl, Antonia
- Leroy, Noé
- Maulu, Marco
- Biay, Sébastien
- Cappe, Zoé
- Konstantinova, Kristina
- Boby, Victor
- Christensen, Kelly
- Pierreville, Corinne
- Aruta, Davide
- Lenzi, Martina
- Le Huëron, Armelle
- Possamaï, Marylène
- Duval, Frédéric
- Mariotti, Violetta
- Morreale, Laura
- Nolibois, Alice
- Foehr-Janssens, Yasmina
- Deleville, Prunelle
- Carnaille, Camille
- Lecomte, Sophie
- Meylan, Aminoel
- Ventura, Simone
- Dugaz, Lucien

## Bias, Risks, and Limitations

The data are skewed toward Old French, Middle Dutch and Spanish, specifically from the 14th century.

The only language that is represented over all centuries is Latin, and in each scripts. The other language with a coverage close to Latin is Old French.

Only one document is available in Old English.

## Citation

**BibTeX:**

```tex
@unpublished{clerice:hal-04453952,
  TITLE = {{CATMuS Medieval: A multilingual large-scale cross-century dataset in Latin script for handwritten text recognition and beyond}},
  AUTHOR = {Cl{\'e}rice, Thibault and Pinche, Ariane and Vlachou-Efstathiou, Malamatenia and Chagu{\'e}, Alix and Camps, Jean-Baptiste and Gille-Levenson, Matthias and Brisville-Fertin, Olivier and Fischer, Franz and Gervers, Michaels and Boutreux, Agn{\`e}s and Manton, Avery and Gabay, Simon and O'Connor, Patricia and Haverals, Wouter and Kestemont, Mike and Vandyck, Caroline and Kiessling, Benjamin},
  URL = {https://inria.hal.science/hal-04453952},
  NOTE = {working paper or preprint},
  YEAR = {2024},
  MONTH = Feb,
  KEYWORDS = {Historical sources ; medieval manuscripts ; Latin scripts ; benchmarking dataset ; multilingual ; handwritten text recognition},
  PDF = {https://inria.hal.science/hal-04453952/file/ICDAR24___CATMUS_Medieval-1.pdf},
  HAL_ID = {hal-04453952},
  HAL_VERSION = {v1},
}
```

**APA:**

> Thibault Clérice, Ariane Pinche, Malamatenia Vlachou-Efstathiou, Alix Chagué, Jean-Baptiste Camps, et al.. CATMuS Medieval: A multilingual large-scale cross-century dataset in Latin script for handwritten text recognition and beyond. 2024. ⟨hal-04453952⟩


## Glossary


![Examples of bookscripts and their name](CATMuS_reference_v2.png)

- Scripts: In the middle ages, the writing style changed over time, specifically in "litterary" manuscripts, for which we call the general scripts "Bookscripts". This is what CATMuS Medieval covers at the time

## Dataset Card Contact

Thibault Clérice (first.last@inria.fr)