pierreq.dubois · PyLaia · Published July 9, 2025
Notaires (7) fin 17e siècle de la Nouvelle-France (1668-1729)
Text Recognition
Description
[English version below]
Modèle développé par le regroupement Les Gardenotes (Québec,CA) (lesgardenotes.org) dans le cadre du projet Nouvelle-France numérique (NFN) : Partenariat collaboratif de gestion des données de recherche (nouvellefrancenumerique.info/).
Le modèle est composé de : jeu d'entraînement de 343 217 mots provenant de: 216 pages d'Antoine Adhémar dit St-Martin, 228 pages de François Genaple, 226 pages de Daniel Normandin, 154 pages de Romain Becquet, 225 pages de Gilles Rageot, 90 pages de Claude Maugue et de 89 pages de Louis Chambalon. Tous les actes choisis ont été transcrits au complet sauf les pages de présentation qui ne contiennent pas assez de texte.
Transcription : Les Gardenotes. Entraînement du modèle : Pierre Dubois. Graphie des notaires : https://lesgardenotes.org/ressources/
[English version]
Model developed by the group “Les Gardenotes” (Quebec,CA) as part of the project "Nouvelle-France numérique (NFN)" : Collaborative partnership for research data management. The model is composed of a training set of 343 217 words from the minute-books of the following seven notaries: Antoine Adhémar dit St-Martin, François Genaple, Daniel Normandin, Romain Becquet, Gilles Rageot, Claude Maugue and Louis Chambalon. All the deeds selected have been transcribed in full except for the presentation pages, which do not contain enough text.
Try this model
Use this modelOpen in Transkribus
Low error rate5.81% CER
Character Error Rate (CER) measures the percentage of characters incorrectly recognised. Lower is better. This model scored 5.81% on its validation set. As a rule of thumb, a CER below 10% is considered good for most handwritten material. This is a larger model trained on diverse material, which generally makes it more robust across different handwriting styles. That said, larger training sets also make it harder to push the CER down further.
Measured on the model's own validation data. Results on your documents may differ depending on handwriting style, document condition, language, and how closely your material resembles the training data.
Words343,217
Lines47,365
Training Pages1,228
Model ID370225