Beyond Comparative Analysis: Making Arguments with Similarity Metrics and Structured Manuscript Data, with a Case Study in Marginal Iconography

Abstract

Abstract: Quantitative measurements of similarity can help researchers understand the increasing amount of manuscript metadata or digital images now available for study. Researchers in archaeology, ecology, and information retrieval have developed diverse methods for quantifying the similarity of categorical, numerical, and presence/absence data. The variety of methods in use reflects the complex associations of the term "similarity" and suggests that quantitative approaches can preserve some of the nuance associated with traditional humanistic approaches. We argue that similarity metrics, contextualized against random simulations, provide flexible tools for identifying and analyzing trends in large sets of multifaceted historical data and can be applied to many kinds of manuscript research projects. Simulations (permutation tests) that highlight how observed patterns diverge from random or theoretical scenarios have allowed us to interpret similarity measurements against a baseline that is analogous to researchers' intuitions born of experience. In this paper, we model using similarity metrics to conduct multifaceted analyses of an iconographic dataset derived from Lilian M. C. Randall's Images in the Margins of Gothic Manuscripts (1966), indexing images in French, English, and Flemish manuscripts produced between about 1250 and 1350. We share examples of approaches such as statistical analysis, Analysis of Similarity, and clustering to model the type of quantitative approaches we believe will become increasingly important for the analysis of digital manuscript data.