Digital Assyriology—Advances in Visual Cuneiform Analysis
Bartosz Bogacz,Hubert Mara
DOI: https://doi.org/10.1145/3491239
2022-05-16
Journal on Computing and Cultural Heritage
Abstract:Cuneiform tablets appertain to the oldest textual artifacts used for more than three millennia and are comparable in amount and relevance to texts written in Latin or ancient Greek. These tablets are typically found in the Middle East and were written by imprinting wedge-shaped impressions into wet clay. There is an increasing demand in the Digital Humanities domain for handwriting recognition, i.e., machine reading of handwritten script, focusing on historic documents. Current practice in text analysis of cuneiform script relies heavily on transliteration and translation, which are incomplete and influenced by the knowledge and experience of the expert that created them. The development of computational tools for cuneiform analysis presents many opportunities. An efficient and accurate sign spotting enables cross-referencing and statistical analyzes that are infeasible to perform manually. Furthermore, a wedge constellation spotting tool, provides experts with a significantly broader base of references to create more accurate and less time consuming transliterations and translations. Yet, cuneiform script has since resisted efforts to computational processing on basis of its basic constituents, its 3D wedge-shaped impressions and their free-form arrangements into signs. In this work, we review the literature on computational processing and recognition in the domain of cuneiform script. We introduce the different heterogeneous sources of cuneiform script, namely, manual ink-on-paper drawings, digital vector graphics drawings, photographs, and 3D scans of tablets. We describe the development of methods, beginning with the first computational classification using a hybrid manual encoding and computational comparison, to the latest methods making use of Generative Adversarial Neural Networks to recognize characters automatically. Finally, we give an overview of applications of these methods that enable quantitative mining in the small, e.g., patterns of wedge constellations, and in the large, e.g., networks of economic activity.