by Every year around 1,500 Latin inscriptions are discovered, which offer an invaluable view of the daily life of the ancient Romans – and a discouraging challenge for historians who are commissioned to interpret.
According to a study published on Wednesday, a new tool for artificial intelligence, which was partly developed by Google researchers, can now help to put together these puzzles from the past.
In the Roman world, the inscriptions in Latin were commonplace, from the determination of the decrees of the emperors to graffiti on the streets of the city. A mosaic outside of a house in the old town of Pompeii even warns: “Caution of the dog”.
These inscriptions are “so valuable for historians because they provide first-hand evidence of old thoughts, language, society and history,” said Yannis Assael, co-author of the study, a researcher at Google’s Ai Lab Deepmind.
“What makes it unique is that the old people themselves are written in all social classes on every topic. It is not just the story written by the elite,” said Aster, who developed the AI model together, to a press conference.
However, these texts have often been damaged for thousands of years.
“We usually don’t know where and when they were written,” said Assael.
The researchers have created a generative neural network that is a AI tool that can be trained to identify complex relationships between the types of data.
They called their model Aeneas after the Trojan hero and son of the Greek goddess Aphrodite.
According to data about the data, locations and meanings of Latin transcriptions, it was trained from an empire that comprised five million square kilometers over two millennia.
Thea Sommerschield, an epigraph at the University of Nottingham, who developed the AI model together, said that “the story is to be studied through inscriptions like solving a gigantic puzzle”.
“You cannot solve the puzzle with a single isolated piece, even though you know information such as its color or shape,” she said.
“To solve the puzzle, you need to use this information to find the parts that connect with it.”
– tested on Augustus –
This can be a big job.
Latin scientists have to compare inscriptions with “potentially hundreds of parallels”, a task that requires “extraordinary scholarship” and “tedious manual search” through massive library and museum collections, according to the magazine in the Nature magazine.
The researchers formed their model on 176,861 inscriptions – worth up to 16 million characters – of five percent of those contained.
It can now estimate the location of an inscription among the 62 Roman provinces, offer a decade when it was produced, and even guess which missing sections could possibly be included, they said.
In order to test their model, the Aeneas team asked to analyze a famous inscription entitled “Res Gestae Divi Augusti”, in which Rome’s first emperor Augustus described his achievements.
The debate between historians when exactly the text was written.
Although the text is interspersed with exaggeration, irrelevant data and incorrect geographical references, the researchers said that Aeneas could use subtle information such as archaic spelling to land on two possible data – the two were discussed between historians.
More than 20 historians who tried the model found that according to Deepmind, it delivered a useful starting point in 90 percent of the cases.
The best results came when historians used the AI model together with their skills as a researcher instead of relying exclusively on one or the other, the study says.
“Generative neuronal networks have seemed in contradiction to educational goals since their breakthrough, since the fear that they rely on AI improve critical thinking rather than knowledge,” said co-authorin Robbe Wulgaert, a Belgian AI researcher.
“By developing Aeneas, we show how this technology can support the humanities sensibly by tackling concrete challenges that historians are confronted.”
Ber-DL/SBK
