Writer Identification and Script Classification. Two Tasks for a Common Understanding of Cultural Heritage
Résumé
Writer identification and script classification are usually considered as two separate and very different tasks, in palaeography as well as in computer science. Following the ICDAR competition on the CLAMM corpus about script classification and dating, this paper proposes to reconsider the tasks and methods of palaeography and computer vision applied to artificial palaeography. We argue that, when aiming at understanding past societies and written cultural heritage in their complexity, palaeography and its core tasks may be defined as discretising the historical and social continuum at different levels of granularity. In this sense, we can consider writer identification and script classification as a single task. We then transfer this hypothesis to computer science, by running two infrastructures created for script classification on a more homogeneous dataset with a focus
on writer identification. The analysis of the results confirms the uniformity of these tasks and allows us to reflect in a novel way on how to demonstrate and illustrate the historical continuum while discretising it in a non-binary way.