1d76ba2084 50187383 inscription grecque

Ithaca AI completes ancient text fragments more accurately than humans

Historians of ancient Greece will now be able to rely on a new tool for studying ancient texts. DeepMind has developed Ithaca, an artificial intelligence capable of filling holes in old, damaged lettering.

You will also be interested

[EN VIDÉO] Interview: how was artificial intelligence born? Artificial intelligence aims to mimic the way the human brain works, or at least its logic, when it comes to decision making. Jean-Claude Heudin, director of the IIM (Institute for Internet and Multimedia) research lab, explains the origins of this study.

British company DeepMind, a subsidiary of Google, continues to phase out its artificial intelligence (AI). This time the company deals with texts in ancient Greek. In an article published in the journal Nature, they present Ithaca, a deep neural network that is helping historians.

Epigraphy is the study of ancient inscriptions made on stone, pottery or metal. This allowed them to survive to our time, but very often the support is damaged or fragments are missing, which makes reading impossible. Historians should then compare with other texts to try to fill in the gaps and establish the date and place of writing.

A tool to help historians

DeepMind, in collaboration with several universities, has taught Ithaca 78,608 inscriptions written in ancient Greek between the 7th century BC. AD and the 5th century AD, as well as from all over the Mediterranean. The AI ​​comes up with multiple solutions for the missing piece, whether it’s a few letters or a few words, with a probability score. It achieves 62% accuracy when restoring text. She manages to determine the geographical origin of the inscription with an accuracy of 71% and the date of its writing within 30 years.

Ithaca is not intended to replace the work of historians, but is an additional tool that must be supervised by specialists. Only experts managed to restore old texts with an accuracy of only 25%, and with the help of AI, this figure increases to 72%. The teams plan to extend the artificial intelligence to other ancient languages ​​such as Akkadian, Demotic, Hebrew and Maya.

Interested in what you just read?