XX Century Italian Handwriting 2.0

This is the second version of a model developed within a project initiated by the Library of the Università Cattolica del Sacro Cuore in Milan to improve access to the documents preserved in its archives. The model includes documents extracted from archival fonds created by ten different 20th-century cultural figures whose archives are held by the UCSC Library. These handwritings, which are chronologically homogeneous and span the entire 20th century, exhibit considerable variability in graphic style, ranging from regular, structured, and looped scripts to large, slanted hands, as well as very small handwriting characterized by frequent variations in letterforms, with many intermediate combinations. To better reflect the composition of the UCSC collections, this second iteration of the model also incorporates documents written by a female author. From a linguistic and lexical perspective, Italian predominates, generally in a medium-to-high register typical of academic and literary contexts. The corpus is further enriched by Latin quotations and passages drawn from both classical and ecclesiastical sources, as well as occasional specialized terminology in English, German, and French. Compared with the first version of the model, XX Century Italian Handwriting 2.0 comprises approximately twice the amount of ground truth data. Although its theoretical Character Error Rate (CER) is slightly higher (7.34% compared to 7.12% for version 1.0), evaluation on a 20-page test set written by previously unseen hands showed improved performance in 75% of the samples. On average, transcription quality increased by 16%, corresponding to a 3% improvement in CER. The project foresees regular updates to incorporate documents from additional 20th-century archival collections held by the University. Trained by Matteo Ruggeri with the contribution of Lorenzo Babini (UCSC Library, Milan).