Artificial intelligence can help predict… the past, as demonstrated by a new technique that is restoring, with great precision, Greek texts from the 5th century BC – reports the magazine Nature this Wednesday (9).
Epigraphy is a branch of history whose objective is to decipher inscriptions on any support, to contribute to the study of ancient civilizations.
Some of these texts have become illegible due to damage to the support (stone, metal…). In addition, these important historical sources have often been moved far from their place of origin, making them difficult to date.
The carbon 14 technique cannot be used on non-organic matter.
A team of researchers from the universities of Venice, Oxford, Athens and the company DeepMind (a subsidiary of Google) have developed a new deep learning tool.
Deep learning is one of the techniques used by artificial intelligence, through algorithms that reproduce the structure and functioning of the human brain.
The new tool is named after Ithaca, in reference to the island of Odysseus in the “Odyssey”. It was perfected after analyzing some 80,000 inscriptions in the Packard Institute for the Humanities database, which has the largest digital collection of ancient Greek inscriptions.
This automatic language treatment technique considers the order in which words appear in the sentence, or the relationship between them, to improve their contextualization.
The program is also capable of taking into account both words and individual characters, distributed in a fragmented way in the support.
A first experience, with a series of decrees from the 5th century BC, carved in stone and found on the Acropolis of Athens, was encouraging.
Ithaca analyzed the texts and managed to suggest, from the historical context that he had previously “studied”, sequences of letters to fill in the missing fragments in the sentences.
Thus, he was able to propose that the word “covenant” be inserted in an oath of obedience of a city in relation to Athens. Thanks to Ithaca, historians were able to decipher the lost texts with up to 72% accuracy, according to Nature.
Ithaca was also able to suggest multiple locations in 84 interconnected regions.
Finally, the tool proposes a precise date for the writing of the decrees: the year 421 BC, that is, 30 years after the approximate dates proposed by historians until today.
“It may sound gimmicky, but it’s essential to understanding classical Greece, in which Pericles and Socrates lived,” Thea Sommerschield of Ca’Foscari University in Venice, co-author of the study, said during an online press conference.
Ithaca can be used for any other ancient language, such as Mayan or Mesopotamian cuneiform, says the expert.