Artificial Intelligence Now Predicts the Past: Ancient Greek Texts from the Vand century BC were able to be restored, dated and located with unparalleled accuracy using a deep learning technique, presented Wednesday in the scientific journal Nature.
Tens of thousands of inscriptions, engraved on stone, clay or metal, allow historians specializing in epigraphy to discover the history of ancient civilizations.
But many have deteriorated over time, to the point where some texts are unreadable as there are missing pieces to the puzzle. These precious sources have also been moved far from their place of origin, which complicates their dating — carbon-14 dating is not possible on inorganic matter.
To help epigraphists decipher them, researchers from the universities of Venice, Oxford and Athens as well as the company DeepMind (Google) have created a deep learning tool, one of the techniques of artificial intelligence that use a “neural network” that mimics the structure of the human brain.
This tool baptized Ithaca, in reference to the island of Odysseus in The Iliad and The Odysseytrained on nearly 80,000 inscriptions referenced in the Packard Humanities Institute database, the largest digital collection of ancient Greek inscriptions.
Its automatic language processing technique takes into account the order in which words appear in sentences and their relationships, in order to better contextualize them.
In view of the multiple gaps in the texts, Ithaca had to integrate both the words and the individual characters, distributed in a fragmentary way on the medium. He then tackled a corpus of decrees of the Vand century before our era, engraved on stone tablets and found on the Acropolis of Athens.
He made hypotheses of letter sequences that could fill in the gaps, consistent with the historical context which he “immersed” himself. For example, he proposed the word “covenant” for six missing characters in a city’s oath of allegiance to Athens.
It is then up to historians to choose the most credible prediction. Their work was greatly facilitated: the Ithaca tool, presented as “accessible”, alone achieved an accuracy of 62%. And when used by historians, its accuracy rate jumps from 25% to 72%, details the study of Naturewhich highlights the beneficial effect of this person-machine “synergy”.
Ithaca further suggests multiple locations in 84 regions, visualized on a map highlighting geographic connections across the ancient world.
Finally, he proposes a precise date for the drafting of the decrees: 421 BC, or about 30 years later than the approximate dates proposed by historians.
“It may seem trivial, but it is essential for our understanding of classical Athens, the time when Pericles and Socrates lived,” said Thea Sommerschield, of Ca’ Foscari University in Venice and co-author of the study, during a press conference.
The Ithaca tool, underlines this epigraphist, can be used for all ancient languages, such as Latin, ancient Maya or cuneiform writing from Mesopotamia.