Every year, approximately 1,500 ancient Latin inscriptions are found within the territory of the former Roman Empire. These texts—carved on stone, painted on walls, or engraved on monuments—provide insight into the lives of everyday Romans. Time has reduced many of them to a damaged and incomplete form, making them hard to interpret.

Fragment of a bronze military diploma from Sardinia, issued by Emperor Trajan to a sailor on a warship, dating to the second century CE. Public domain via the Metropolitan Museum of Art.

Now, researchers at Google DeepMind and their academic collaborators have developed Aeneas, an artificial intelligence tool named after the Trojan hero of Roman mythology. Aeneas can restore missing words, detect the date and place of writing, and seek parallels with other inscriptions within seconds.

Aeneas was trained on over 176,000 inscriptions from vast epigraphic databases such as EDR (Epigraphic Database Roma), Heidelberg, and Clauss-Slaby. That content includes both text and images and spans over 16 million characters. Aeneas doesn’t simply search for similar phrases—instead, it identifies historical patterns across time and geography.

“What makes them unique is that they are written by the ancient people themselves, across all social classes, on any subject,” said Yannis ᴀssael, a Google DeepMind researcher and co-author of the study. “They’re not filtered by elite historians; they’re the real voices of the past.”

Drawing of the Res Gestae Divi Augusti. Public domain

But many of these inscriptions are fragmentary, and their definitive origins are uncertain. According to Thea Sommerschield, an epigrapher at the University of Nottingham who co-developed the AI model, “Studying history through inscriptions is like solving a gigantic jigsaw puzzle. You can’t solve the puzzle with a single isolated piece, even though you know information like its color or its shape.”

Aeneas helps by comparing a new inscription with thousands of known examples and suggesting possible reconstructions. The model can recover missing Latin text with up to 73% accuracy when the gap is limited to ten characters. When the length is unknown, accuracy drops to 58%. It can also ᴀssign texts to one of the 62 provinces with 72% accuracy and estimate dates within approximately 13 years.

In one instance, Aeneas ᴀssessed the famous Res Gestae Divi Augusti, a victory inscription wherein Emperor Augustus listed his achievements. Historians continue to debate the date of the work. Aeneas suggested two probable date ranges—10 to 1 BCE and 10 to 20 CE—which match prevailing scholarly views.

The tool was found to be useful for historians. In a cooperative test, 23 experts utilized Aeneas to examine inscriptions and discovered its recommendations useful in 90% of cases. But although Aeneas offers powerful support, experts point out that it is no replacement for human expertise.

Aeneas builds on an earlier Google project, Ithaca, designed to aid in the interpretation of ancient Greek inscriptions. As with its predecessor, Aeneas is available to anyone: there’s an interactive version at predictingthepast.com, and the code and dataset have been open-sourced.

More information: ᴀssael, Y., Sommerschield, T., Cooley, A. et al. (2025). Contextualizing ancient texts with generative neural networks. Nature. doi:10.1038/s41586-025-09292-5