A New A.I. Can Help Historians Decipher Damaged Ancient Greek Texts

The developers announced that their program, called “Ithaca,” is able to reconstruct missing or damaged inscriptions

Stone Tablet with ancient greek texts
Many surviving ancient descriptions are fragmented and unable to be radiocarbon dated. Now researchers say a new artificial intelligence program may help fill in those gaps. Public domain via Wikimedia Commons under CC BY-SA 4.0

A new artificial intelligence program (A.I.) could help fill in gaps in ancient Greek texts and determine when and where they’re from, according to the Guardian’s Nicola Davis. In a paper published in Nature, the program’s developers announce that their new A.I. system can reconstruct portions of missing or damaged inscriptions with up to between 62 to 72 percent accuracy, depending on data input.

Nicknamed “Ithaca” after the Greek island home of King Odysseus, the A.I. was developed by Alphabet subsidiary DeepMind. To test the system, the team used data from more than 60,000 well-studied ancient Greek texts, dated between 700 B.C. and 500 C.E., reports Carissa Wong for New Scientist. By obscuring some of the texts, the scientists could determine how accurately Ithaca filled in the missing portions.

Without assistance, Ithaca was able to restore texts with 62 percent accuracy; determine a text’s geographic origins 71 percent of the time; and date texts to within 30 years from when it was written, according to the study. Historians working without benefit of the program could restore texts with only 25 percent accuracy. But when historians supplemented data from the A.I. system and added it to their own work, historical accuracy increased to 72 percent.

The researchers believe Ithaca will prove an invaluable tool to historians in reconstructing ancient texts. Study co-author Thea Sommerschield, of Ca’ Foscari University of Venice and Harvard University, tells the Guardian that most of these surviving texts are “fragmentary or illegible,” and may have been moved from their original location. Stone inscriptions can’t be radiocarbon dated, making it more difficult to determine their age.

Ithaca has already been used to help date inscriptions discovered in the Acropolis of Athens, and Sommerschield says the Ithaca software has even broader potential. Though Ithaca was trained on Greek inscriptions, the software’s flexibility would allow it to be configured to any number of ancient writing forms, from Mayan to cuneiform.

“Just as microscopes and telescopes have extended the range of what scientists can do today, Ithaca aims to singularly augment and expand the capabilities to study one of the most significant periods of human history,” study co-author Yannis Assael tells the Guardian.

University of Reading classicist Eleanor Dickey tells the Verge’s James Vincent she’s excited to try out the software—which is free and available online—but is skeptical of the new system, noting that “when people rely on it they will need to keep in mind that it is wrong about one third of the time.”

“It is not yet clear to what extent use of this tool by genuinely qualified editors would result in an improvement in the editions generally available—but it will be interesting to find out,” Dickey says.