Language Log

« previous post | next post »

AI Is Helping Historians With Their Latin
A new tool fills in missing portions of ancient inscriptions from the Roman Empire

By Nidhi Subbaraman Aug. 6, 2025

In recent years, we have encountered many cases of AI assisting (or not) in the decipherment of ancient manuscripts in diverse languages.  See several cases listed in the "Selected readings".  Now it's Latin's turn to benefit from the ministrations of artificial intelligence.

People across the Roman Empire wrote poetry, kept business accounts and described their conquests and ambitions in inscriptions on pots, plaques and walls.

The surviving text gives historians a rare glimpse of life in those times—but most of the objects are broken or worn.

“It’s like trying to solve a gigantic jigsaw puzzle, only there is tens of thousands more pieces to that puzzle, and about 90% of them are missing,” said Thea Sommerschield, a historian at the University of Nottingham.

Now, artificial intelligence is filling in the blanks.

An AI tool designed by Sommerschield and other European scientists can predict the missing text of partially degraded Latin inscriptions made hundreds of years ago and help historians estimate their date and place of origin.

Like many folks nowadays, Sommerschield and colleagues personalize their AI helper:

The tool, called Aeneas, was trained against a database of more than 176,000 known Latin inscriptions created over 1,500 years in an area stretching from modern-day Portugal to Afghanistan, said Yannis Assael, a staff research scientist at Google DeepMind who was part of the project team.

People used the Latin language differently depending on where and when they lived. This adds to the challenge of pinpointing the meaning and provenance of found inscriptions, but it also presents clues that historians can use.

How does Aeneas work its magic?

Aeneas compares a given sequence of letters against those in its database, bringing up those that are most similar, essentially automating at a massive scale what historians would do manually to analyze a newly found artifact.

Nearly two dozen historians who tested the tool found it helpful 90% of the time, the team that developed it reported in the journal Nature in July.

Because Aeneas works best where there are many known inscriptions from a given place and time, it may be of less help if something truly unique turned up, said Anne Rogerson, a Latin scholar at the University of Sydney who wasn’t involved with the work.

"But most inscriptions are quite formulaic, so this isn’t going to be an issue a lot of the time,” she said.

Aeneas is highly versatile, being applicable to a wide variety of texts, from the mundane to the sacerdotal.

Among the tests, the team deployed the tool on the text of a famous Roman inscription on the walls of a temple in modern-day Ankara, Turkey. Called the Res Gestae Divi Augusti, it describes the reign of the Roman emperor Augustus.

In its analysis Aeneas offered two likely time spans for when the inscription was made, mirroring an existing debate among historians who are split over whether the text was created during Augustus’s lifetime, or after his death.

“Bang on,” said Sommerschield. “It shows how tools like Aeneas can be used for modeling historical uncertainty.”

When scholars rely substantially on AI tools for their successful reconstructions and decipherments, they should list the collaborators among the contributors.

 

Selected readings

• "AI (and human ingenuity) to the rescue" (2/6/24)
• "Decipherment of Linear A" (5/24/23)
• "Toward the decipherment of Harappan" (2/14/22)
• "Hype over AI and Classical Chinese / Literary Sinitic" (11/9/23)
• "Digitization of Babylonian fragments" (2/12/23)
• "Decipherment of the Indus script: new angles and approaches" (3/6/25)

[h.t. François Lang]

August 9, 2025 @ 6:38 am · Filed by Victor Mair under Artificial intelligence, Decipherment

Permalink