Researchers at Google DeepMind and Oxford have created an AI system to recover fragmented words and messages written in ancient texts.

The tool, named Pythia, was trained on a large corpus of ancient Greek inscriptions. It can be used to to decipher damaged writing on stone or scrolls, researchers say, and assist historians in their restoration work.

The tool, researchers say, is possibly the first of its kind:

This work presents a novel assistive method for providing text restorations using deep neural networks. To the best of our knowledge, Pythia is the first ancient text restoration model that recovers missing characters from a damaged text input.

When tasked with deciphering damaged text during training, Pythia completed texts with a 30.1% character error rate, compared with human rate of 57.3%. While not nearly a wholly reliable tool, it is nonetheless state of the art.

Pythia takes its name from the priestess of Apollo at Delphi, in Greek myth.