Automated Cuneiform Symbol Detection and Translation Using Deep Learning Techniques

This paper introduces an automated methodology for the detection and translation of cuneiform symbols using high-level deep learning techniques. A total of five deep learning models were trained on a comprehensive dataset comprising cuneiform characters, which were evaluated according to major metrics such as precision and accuracy. After a thorough evaluation, two models demonstrated superior results and were further tested on cuneiform symbols extracted from the Hammurabi law corpus. Eventually, each model identified the Akkadian representation of each cuneiform symbol and provided its English translation. To further optimize the performance, ensemble and stacking techniques will be used in future work, combining these top models by creating hybrid architectures that improve the overall accuracy at detection. In addition to model development, this paper also explores the linguistic relations shared by Akkadian, an ancient Mesopotamian language, with Arabic, hence identifying the historical linguistic influences. This research presents an interdisciplinary investigation that bridges computational linguistics and archaeology, thus showing an emerging capability of deep learning in deciphering ancient scripts, which will contribute significantly to historical and cultural studies.