top of page
  • Voltaire Staff

Researchers use AI to decipher 5,000-year old cuneiform writings

Image Courtesy: Unsplash

A group of scientists from Germany has made a computer programme that can understand cuneiform writings from Mesopotamian civilization.

The clay tablets researchers collected to study carry mysterious sign and symbols etched on them and are about 5,000 years old. The researchers used a special technology to create 3D models of these tablets and teach the AI to read them.

Mesopotamia in modern-day Iraq is known as one of the first places where humans lived in cities. These tablets have ancient languages like Sumerian, Assyrian, and Akkadian written on them.

Hubert Mara, one of the people who worked on the study, said, "You can find everything on them: from what people bought to decisions made in courts." But because they're very old and damaged by weather, even experts find the content on the cuneiform hard to read and understand.

Published in the Eurographics Association Journal, the study involved the use an AI model called Region-based Convolutional Neural Network (R-CNN), built specifically for recognizing symbols on these tablets. The researchers studied nearly 2,000 3D tablet models and detailed more than 21,000 symbols and 4,700 wedges.

The process had two main parts: first, it detected the symbols using a model called RepPoints combined with ResNet18, which helps recognize patterns in the images related to the ancient languages. Then, another tool called Point R-CNN, along with advanced features like Feature Pyramid Network and RoI Align, identified and predicted the positions of the wedges.

The wedges are the basic elements of the cuneiform writing, essentially allowing the AI to 'read' the text through a 3D rendering.

By using 3D scans and detailed measurements of things like the depth of the marks made on clay and the spaces between symbols, this method overcame issues faced by traditional 2D photos, like inconsistent lighting and colours. This helped the AI analyze the ancient writings more accurately.

Typically, when studying ancient writings, researchers use optical character recognition software, or OCR, to change scanned images or 2D photos of the text into readable text for machines, researchers said.

Ernst Stötzner, a co-author, said that OCR usually deals with paper or parchment writings without much trouble. However, when it comes to cuneiform tablets, it becomes trickier because factors like lighting and viewing angles significantly affect how well certain characters can be identified.

According to researchers, all from Martin Luther University Halle-Wittenberg, Johannes Gutenberg University Mainz, and Mainz University of Applied Sciences, a total of twelve cuneiform languages are known to exist to date.

In the future, they believe, their software could also help decipher such inscriptions as found in cemeteries, which are three-dimensional like the cuneiform script.


bottom of page