Comparing Arabic Handwriting Styles Using Conventional and Computational Methods

Kyle Ann Huskin

Traditional palaeographic analysis relies on visual examinations of handwriting samples – from a script’s ‘general impression’ to single-grapheme comparisons – in order to answer important questions about a written artefact, such as: Did one person or multiple people write the object? Can the writer’s identity be determined? Can an approximate date for the artefact be established? Can multiple artefacts be linked together by the presence of the same writer’s script? 

Conclusions drawn from traditional methods can now be corroborated and strengthened by computational pattern classification methods. A script-independent and learning-free software, the Handwriting Analysis Tool (HAT-2) uses ‘keypoint detection’ algorithms to locate salient image features and assign their features a numerical ‘descriptor’ based on the vector magnitudes and orientations from every position in the keypoint; from there, ‘similarity scores’ for different handwriting samples are calculated based on these descriptors (Mohammed et al. 2020, 80–83). As the name suggests, the similarity scores give scholars an indication of which hands are most alike.

This information can be used in a variety of ways, including to link different texts written by the same person, as in Mohammed et al. 2020. In this case, the same text from three different ‘audience certifications’ (Arabic samā‘āt) in two codicological units of the same manuscript (Forschungsbibliothek Gotha, Ms. orient. A 627, fols 13b–15b and 37b) was analysed with HAT-2. Seidensticker 2015 determined with traditional palaeographical methods that all three texts were written by the same hand. The results of HAT-2 analysis confirm this conclusion, showing that all three texts have a much higher similarity score to one another than to any of the ten other audience certification texts in the manuscript (Mohammed et al. 2020, 85).