Scientists from the University of Chicago have developed a machine learning system that can automatically translate texts on ancient clay tablets.

According to The University of Chicago News, the DeepScribe system will initially be used to decrypt cuneiform records, which were used in the ancient Iranian Achaemenid empire (550-330 BC).

Existing computer systems experience certain difficulties in translating such texts due to the complexity of the characters and the three-dimensional shape of the plates on which they are written. According to researchers from the University of Chicago, their system is able to cope with the task.

To create a model, more than 6000 annotated images of texts of that time are used as a “simulator”. Their full decryption will provide information on the history, society and language of the Achaemenids. The training is based on the Achaemenid language dictionary, which contains more than 100,000 individual characters.

Sanjay Krishnan, a professor of computer science at the University of Chicago, used this annotated set to teach AI how to read other tablets that he still did not know. As a result, the system was able to decrypt the characters contained in them with 80% accuracy.