Henri Pirenne Manuscripts Dataset

This dataset contains manually corrected ground truth data for Handwritten Text Recognition (HTR) of manuscripts written by the Belgian historian Henri Pirenne. The manuscripts date from 1909 to 1935 and were produced in Pirenne’s handwriting. The dataset was created using four manuscripts listed below. Title Date Number of pages Mahomet et Charlemagne 1935 215 Les anciennes démocraties des Pays-Bas 1909 236 Préparation de « Mahomet et Charlemagne » 1922 13 Journal de guerre 1916 9 Total 473 Three of these manuscripts are preserved in the Pirenne collection held by the Archives, Heritage and Special Collections of the Université Libre de Bruxelles. The Journal de guerre manuscript is preserved at the library of Ghent University. Layout Description For the description and annotation of the manuscripts, we adopted the controlled vocabulary SegmOnto. The main segmentation zones used in the dataset are: MainZone, NumberingZone, TitlePageZone, MarginTextZone. Text lines were annotated using the following categories: DefaultLine, HeadingLine, InterlinearLine. The SegmOnto ontology also allows the definition of sub-zones and line subcategories. For example, the following specialized categories were used: MarginTextZone:lower for footnotes, CustomZone:numbering for page numbering of unknown origin, InterlinearLine:addition for interlinear additions, InterlinearLine:correction for corrections written after crossed-out words. Dataset Content This dataset archive contains fully manually corrected transcriptions and layout annotations of 473 manuscript pages from the manuscripts listed above. All files are provided in ALTO-XML format. The manuscript images are not included in the dataset. For access to the images, please contact the respective holding institutions. Dataset Statistics Text unit Count Pages 473 Lines 20,757 Regions 34,231 Characters 797,311