AI Tool Deciphers Ancient Roman Inscriptions in Seconds, Transforming Historical Research

Google DeepMind released an AI system that analyzes ancient Roman inscriptions in seconds—work that normally takes historians days or weeks. The system, called Aeneas, examines Latin text carved into stone monuments, tomb markers, and everyday objects from across the Roman Empire.

Historians spend enormous time hunting through databases and books to find similar inscriptions that help them understand damaged or incomplete text. Aeneas does this automatically, scanning 176,000 Latin inscriptions to find relevant parallels and fill in missing words.

The research team, led by Google DeepMind scientists and University of Nottingham historians, published their findings in Nature this week. They tested Aeneas with 23 expert historians who regularly work with inscriptions. Historians became more accurate and confident when using the AI tool.

Finding needles in historical haystacks

Romans carved words into everything. Emperor proclamations, love poems, business deals, tombstones—you name it. Archaeologists find about 1,500 new inscriptions every year. Most are broken, weathered chunks of stone with words missing.

"Writing was everywhere in the Roman world," says Thea Sommerschield, who studies ancient texts at the University of Nottingham. She helped build Aeneas. The inscriptions show how regular people lived, not just what emperors wanted everyone to think.

Historians search for "parallels"—other inscriptions with similar wording, formulas, or origins. These connections help them figure out what missing text might say, where an inscription came from, and when it was made. Tracking down these parallels requires deep expertise and access to vast libraries.

Training an AI on Roman stones

Aeneas changes this. The system learned patterns from a massive database called the Latin Epigraphic Dataset, which combines three major inscription collections. Historians had built three separate databases over the decades. Each used its own format, so combining them required tedious cleanup work. The team standardized everything into one searchable collection.

Aeneas analyzes writing style, not just words. It notices patterns in how Romans structured sentences and used certain phrases.

Aeneas can restore missing text, date inscriptions, and determine where they came from. The key advance is working with gaps of unknown length. Previous systems needed to know exactly how many letters were missing.

Putting historians to the test

To measure Aeneas's impact, the research team ran the largest "historian and AI" study ever conducted. Twenty-three experts worked on inscription puzzles both with and without AI assistance.

When historians used Aeneas's suggested parallels, their accuracy jumped 35% for determining where inscriptions came from. With the AI's predictions included, accuracy shot up 152%. For dating inscriptions, historians got 32% closer to the correct time periods when using Aeneas.

"The parallels retrieved by Aeneas completely changed my perception of the inscription," one historian wrote in feedback. "It would have taken me a couple of days rather than 15 minutes to find these texts."

Solving ancient mysteries

The researchers tried Aeneas on the Res Gestae Divi Augusti—basically Augustus bragging about everything he accomplished as emperor. Historians argue about when he actually wrote it.

Aeneas looked at the language and came up with two date ranges: 10-1 BCE and 10-20 CE. That's exactly what scholars have been arguing about, which shows the AI gets the uncertainty right.

The system also revealed something new. It found that the inscription shares subtle language patterns with Roman legal documents and imperial messaging—connections human historians hadn't previously noted.

In another test, Aeneas analyzed a damaged altar from ancient Mainz, Germany. The AI correctly identified its most relevant parallel: another altar found nearby that used nearly identical rare formulas. This connection helped historians understand how religious practices spread through the Roman provinces.

Processing text and images

Aeneas processes both text and images. It can look at photos of inscriptions and factor in visual elements like stone shape and layout. This multimodal approach helps determine where inscriptions came from with 72% accuracy across 62 Roman provinces.

The system dates texts by noticing how Romans wrote differently over time. Spellings changed, job titles evolved, and they mentioned different buildings or events depending on the era.

Mary Beard from Cambridge thinks Aeneas could change how historians work. She called it "really exciting" and expects it will reshape the field.

Free tools for everyone

The research team made Aeneas freely available online at predictingthepast.com. Historians, students, teachers, and museum workers can use it to analyze Latin inscriptions. The team also open-sourced the code and dataset to support further research.

They're already working with schools in Belgium to integrate Aeneas into history education. The tool aligns with European digital literacy frameworks and helps students engage directly with ancient sources.

The system can be adapted to other ancient languages and texts—from Greek papyri to coins and manuscripts. This could transform how historians work with fragmentary evidence across many civilizations.

Why this matters:

• Speed wins: What took historians days of digging through books now happens in seconds, so they can spend time figuring out what the texts actually mean.

• Anyone can play: Small colleges and students get the same research tools that only big universities had before.