Session 8: KI in den Geschichtswissenschaften II – Hybride KI und Semantic Web
|Datum & Zeit:
16:40 – 17:40 Uhr
|Auditorium des Jacob-und-Wilhelm-Grimm-Zentrums
Are domain-specific theoretical approaches valuable for the application of new computational methods? The case study of Erwin Panofsky’s artworks interpretations and the semantic web
Sofia Baroncini, Marilena Daquino, Francesca Tomasi
University of Bologna, Italien
Recently, researchers have shown an increased interest in the analysis of art-related topics with computational methods. As some studies highlight, approaches that integrate traditional theoretical perspectives in the new computational methods may have some benefits, including the application of an established approach to a wider number of artworks.
In this talk, we address the possibilities offered by the expression of an art historian’s method in an ontological perspective through a case study on Erwin Panofsky’s approach and work. To this end, we created a Linked Open Data dataset containing interpretations mainly by the art historian about ca. 400 artworks, mostly from Middle Ages and Renaissance Western art, modelled according to a newly-created ontology based on the art historian’s theory. The research aims at verifying 1) if data structured according to Panofsky’s theory allow answering to domain research questions, and 2) if characteristics emerging from the data analysis correspond to the description of his theory.
Results show the creation of an ontology and semantic web data following Panofsky’s theory can be a good tool for applying a computational perspective to the studies of the domain of iconography and iconology and to quantitatively characterize the art historian’s approach. Additionally, we show that other computational applications, such as computer vision, can benefit from semantic integration with data from an established tradition to perform better results.
‘Hybrid AI’ as an aid to interpret visual sources? The application of semantic web technologies and machine learning to analyse heraldic communication in pre-modern manuscripts
Humboldt-Universität zu Berlin
The aim of this paper is to discuss the use of hybrid AI as an aid to interpret historical visual sources. Hybrid AI is understood as a combination of knowledge graphs (symbolic AI) and machine learning methods (sub-symbolic AI) as part of a single architecture. Such systems have not yet been used in historical studies, apart from a few exceptions.
The paper offers a case study in which the visual communication of hierarchies by means of specifically arranged collections of coats of arms in medieval manuscripts is investigated. Explicitly modelled knowledge about individual coats of arms and their contextualisation is brought together with visual information generated by the automatic indexing of a large amount of image data.