Voices of the People, Aarhus University · Fields · Published November 16, 2025

Field-model, 1700-tallets supplikprotokoller, supplik og svar

Field ExtractionScholar+

Description

Modellen er trænet på håndskrevne supplikprotokoller fra Danske Kancelli fra 1740’erne til 1760’erne. Den er trænet til at adskille supplikker i højre spalte og sagsbehandling og svar i venstre side af protokoller, og den tagger begge spalter med hhv. ‘supplik’ og ’svar’. Den adskiller desuden de enkelte supplikker og tilhørende svar i felter. Den nummererer læserækkefølgen på felterne, startende med den øverste supplik i højre spalte efterfulgt af svaret til denne supplik. Modellen er trænet af Line Keller Nørbøge Ottosen som en del af projektet Voices of the People, ledet af Nina Javette Koefoed og støttet af Carlsbergfondet. Læs mere om projektet bag på https://cas.au.dk/voices. The model has been trained using petition protocols in Kurrent/Gothic handwriting from the Danish central administration (Danske Kancelli), dating from the 1740s to the 1760s. It is trained to distinguish between petitions in the right-hand column and answers in the left-hand column. Both columns are labelled 'supplik' (petition) or 'svar' (answer), respectively. The model also separates each petition and its adjacent answer into fields and applies numbers to them, starting with the first petition in the right-hand column of a page and moving on to its answer in the left-hand column. Line Keller Nørbøge Ottosen trained the model as part of the Voices of the People project, which is led by Nina Javette Koefoed and funded by the Carlsberg Foundation. Further information can be found at: https://cas.au.dk/voices.
Field-model, 1700-tallets supplikprotokoller, supplik og svar
Open in Transkribus
Good precision56.44% MaP

Mean Average Precision (MaP) measures how accurately the model detects field regions (higher is better). This model scored 56.44% on its validation set. MaP is harder to compare across models than CER, because the score depends heavily on how many distinct region types the model must distinguish. A model detecting a handful of simple fields will naturally score higher than one trained to recognise many fine-grained regions, even if both perform well in practice.

This score reflects performance on the model's own validation data. Your results will depend on how closely your documents match the training material and the complexity of the structures you need to detect.

Words896,858
Lines151,678
Training Pages1,063
Model ID433805