Paleography Powered by Learning: Analyzing Cuneiforms with Neural Networks

Cuneiform, the world’s oldest writing system, was invented in ancient Mesopotamia around 3200 BCE, and was in use for over three thousand years throughout the ancient Near East.

The known corpus of cuneiform documents is large and diverse, comprising an estimated 500,000 clay tablets inscribed in various languages and genres. However, dating these texts and parsing their contents currently requires painstaking work from experts, a process that is neither scalable nor amenable to digital analysis.

In our research collaboration, we aim to build a digital paleographic pipeline using recent developments in neural networks. In particular, we are interested in leveraging powerful text-to-image generative models, and modifying these for understanding (and generating) cuneiform.