AI model Aeneas supplements fragmented Latin inscriptions
The AI Aeneas can supplement ancient inscriptions preserved in fragments and place them chronologically. Together with historians, it produces better results.
One of the Latin inscriptions added by Aeneas.
(Image: The Metropolitan Museum of Art/Public Domain)
A team of scientists from the University of Nottingham and Google DeepMind have developed the AI model Aeneas to supplement and chronologically classify partially preserved ancient Latin inscriptions. However, artificial intelligence (AI) is only intended as a tool to help historians find missing fragments out of context in order to better decipher ancient inscriptions.
The number of Latin inscriptions discovered by archaeologists is constantly increasing. They are often carved in stone, ceramic or metal, but are weathered and only preserved in fragments, which makes them difficult to translate and interpret. The AI Aeneas, developed by researchers at the University of Nottingham and Google Deepmind, is able to analyze the fragments of writing and add to them based on the historical context. The missing sections are predicted. The AI is named after the Trojan prince Aeneas from Homer's Iliad, who is considered the bravest Trojan after Hector, who fought against Achilles.
Temporal classification and additions to Latin inscriptions
The scientists developed a generative neural network that can recognize the complex relationships between information. The researchers describe the AI in the study "Contextualizing ancient texts with generative neural networks", which has been published in Nature. The model was trained with dates, locations and meanings of Latin inscriptions from the entire Roman Empire between 800 BC and 700 AD. A total of 176,861 Latin inscriptions were used for the training. Around 5 percent of these also contained image information. The AI model trained in this way is able to predict the origin of a new inscription from 62 Roman provinces and assign it to a year of origin. This works on average with a deviation of 13 years.
The researchers had Aeneas analyze the "Res Gestae Divi Augusti" (The Deeds of the Deified Augustus), a fragmentary inscription that has been preserved as a copy in three cities of the Roman Empire. In the inscription, the first emperor of the Roman Empire, Augustus, gives a kind of performance and account of his reign. The text is riddled with exaggerations, unclear dates and geographical inaccuracies. In addition, archaic spellings were used, making it difficult to place the inscription in chronological order. However, Aeneas was able to narrow down the text to the time periods that are currently being discussed in research.
Videos by heise
Aeneas can also supplement lost parts of inscriptions of any length. To do this, the AI uses textual and contextual parallels. The additions made by Aeneas can then be used by historians as a starting point for their research. The study showed that 90 percent of the historians involved found the AI's advice useful. The collaboration between humans and AI produced better results than either humans or AI would have been able to achieve on their own.
(olb)
