Is medieval distant viewing possible? : Extending and enriching annotation of legacy image collections using visual analytics.

This article focuses on developing a visual analytics system to unify and enhance the annotation and hierarchical classification of medieval manuscript images from two legacy iconographic databases, Mandragore and Initiale, which contain overlapping but inconsistent metadata. The system integrates machine learning techniques, such as image and word embeddings, with interactive visualizations to assist domain experts—medievalists, paleographers, and art historians—in reconciling vocabularies, exploring image similarities, and creating a structured label hierarchy tailored to the specific cultural and iconographic context of 13th- and 14th-century Latin bibles. By categorizing labels into descriptive, decorative, and interpretive types, the approach addresses challenges unique to historical image datasets that differ from contemporary image hierarchies, facilitating improved metadata interoperability and supporting future supervised machine learning tasks. The authors present qualitative evaluations, usage scenarios, and discuss limitations and future directions, emphasizing the system’s adaptability to other cultural heritage collections facing similar issues of siloed and inconsistent metadata.