Work In Progress

!soon adding a webpage !

_{Malamatenia Vlachou Efstathiou}

Abstract

This study introduces an interpretable scribal hand characterization and variability analysis based on a deep learning framework combining graphic tools with statistical analysis of scribal practices. Designed to bridge the gap between traditional palaeographic methods based on qualitative observations and automatic computational models, our approach enables interpretable character-level inter- and intra-scribal variation analysis and graphic profiling. We demonstrate our method to Charles Vs copy of the Grandes Chroniques de France (Paris, BnF, fr. 2813), revisiting the traditional attribution to two royal scribes of Charles V, Henri de Trvou and Raoulet dOrlans. Through the definition of graphic profiles and complementary statistical analysis of abbreviation usage and space management, we offer a more systematic and context-aware understanding of scribal behaviour. Beyond this case study, the approach opens new possibilities for palaeographic inquiry, from mapping script evolution to characterizing scribal variability in a communicable and interpretable way.

Repository Structure

• In the data folder, the data is available in two forms, ground truth (images + XML ALTO files) and the processed dataset for our experiment. Data are available also on Zenodo: DOI:10.5281/zenodo.15282371

data/ raw_ground_truth/ # Ground Truth (GT) from selected folios of the Paris, BnF, fr. 2813: Images + ALTO XML files with graphemic transcription and layout tagging (SegmOnto) processed_dataset/ # Dataset curated for our analysis using the Learnable Handwriter

• In the scripts folder, we include the notebooks and utils for the analysis notebooks/ utils/ # folder contains functions for the notebooks filter_sprites.ipynb pca.ipynb statistical_analysis.ipynb

• In the results folder, we include the

prototypes/ # Letter prototypes cropped/ filtered/ finetuned/ ... transcribe.json # Mapping between characters and their indices prototypes_paper_grid.jpeg # Overview of the prototypes

Prototype Generation:

Character prototypes are generated using the Learnable Typewriter approach. The Learnable Typewriter is a deep instance segmentation model designed to reconstruct text lines by learning the dictionary of visual patterns that make it up. Given an input image of a text line, the models task is to reconstruct the input image, by compositing the learned character prototypes onto a simple background. Each prototype is a grayscale image can be thought of as the optimized average shape of all occurrences of a character in the training data, standardized for size, position, and color. Training the model on a specific corpus such as manuscripts in a particular script type or a particular hand produces a set of ideal letterforms of the given corpus, resembling the abstracted alphabets used for palaeographical analysis. It has been adapted to handle medieval handwriting and prototype comparison in the Learnable Handwriter version. If you want to learn more about the prototype generation and train the Learnable Handwriter on your data, please refer to the tutorial page.

• as well as the results of the notebooks ``` graphicprofilespca/ prototypes/ statistical anlaysis/
  

```

Run the analysis on our dataset and reproduce the paper's results

You can either clone the repository or run directly the notebooks in Google Colab using the following links:

• filterprototypes.ipynb[](https://colab.research.google.com/github/malamatenia/palaeographic-variability-analysis-grandes-chroniques-fr-2813/blob/a0ca27a7a03f2474849d0e893f1c13c10de8d907/scripts/filterprototypes.ipynb)
• pca.ipynb
• statisticalanalysis.ipynb[](https://colab.research.google.com/github/malamatenia/palaeographic-variability-analysis-grandes-chroniques-fr-2813/blob/a0ca27a7a03f2474849d0e893f1c13c10de8d907/scripts/statisticalanalysis.ipynb)

Cite us (article tba soon)

bibtex @misc{vlachou2025variability, title = {Interpretable Deep Learning for Palaeographic Variability Analysis; revisiting the scribal hands of Charles V Grandes Chroniques de France (Paris, BnF, fr., 2813)}, author = {Vlachou-Efstathiou, Malamatenia}, year = {2025},

Acknowledgements

This study was supported by the CNRS through MITI and the 80|Prime program (CrEMe Caractrisation des critures mdivales), and by the European Research Council (ERC project DISCOVER, number 101076028). I would like to express my deepest gratitude to my advisors, Prof. Dr. Dominique Stutzmann (IRHT-CNRS) and Prof. Dr. Mathieu Aubry (IMAGINE-ENPC), whose guidance, insightful feedback, and proofreading, as well as continuous support, were instrumental throughout the writing of this paper.

Check out also our other projects: - Vlachou-Efstathiou, M., Siglidis, I., Stutzmann, D. & Aubry, M. (2024). An Interpretable Deep Learning Approach for Morphological Script Type Analysis. - Siglidis, I., Gonthier, N., Gaubil, J., Monnier, T., & Aubry, M. (2023). The Learnable Typewriter: A Generative Approach to Text Analysis.