Samara Petros (University of Amsterdam / Huygens ING), He, Sheng, Msc, University of Groningen
Dating and localizing undated medieval manuscripts can be a daunting task, and can often only be achieved on the basis of certain handwriting characteristics. Usually, however, the dating of a script is not based on objective criteria but on the individual non-verbal intuition of paleographic experts, who often arrive at different conclusions. This state of affairs is highly unsatisfying, but not surprising, because there is a notorious lack of a sufficiently large and chronologically complete collection of dated manuscripts to be used as a reference corpus.
The purpose of this paper is to provide an outline of how to construct an objective palaeograpical ‘scale’ of datable elements in late medieval handwriting (1300-1550), which can be used as a more or less objective aid or ‘guide’ to date and also localize undated manuscripts. This scale will be based on material that hitherto was neglected for this purpose: charters and other documentary sources in the (city) archives, material that is generally precisely dated and localized – and available in abundant quantities. These administrative documents were often written by the same scribes who wrote the undated manuscripts, using the same types of script.
The method brings two domains of expertise: a palaeographer making a careful selection of documents from several different city archives, and analyzing their handwriting, and, in close co-operation, an expert from the field of pattern recognition and machine learning constructing algorithms that also will be able to estimate the date of a handwritten specimen on the basis of training on a reference data set and testing on known and unknown samples.
Already, there are various computer programs for automatic writer identification, the most promising of which is in this context surely the Groningen Intelligent Writer Identification System (GIWIS; Brink et al., 2011), recently developed at the University of Groningen, which has already proven that it is quite capable of handling medieval script. The GIWIS approach makes use of several ‘shape feature’ groups, each tapping into a particularity of a script type: textural, allographic, and ink-trace formation oriented, the latter feature group focusing specifically on the characteristics of manuscripts written with a quill. If it is demonstrated that the computer can classify specimens of that script according to individual scribal features, it is reasonable to expect that more general shape classes such as those which are characteristic for the script of a certain period and/or region can be automatically detected.
Whereas the broad knowledge of the human palaeographer is essential, computers can work much faster, thus facilitating a rapid first selection of huge quantities of digitized manuscript images. In the end, the proposed palaeographical scale will make the task of dating medieval manuscripts more objective, much quicker, and less dependent on the individual expert. The scale can be used instantly by countless medievalists in the Netherlands and abroad, thus facilitating the scholarly use of thousands of medieval manuscripts.