Machine learning techniques are giving archaeologists new tools to help them understand the past, particularly when it comes to reading ancient texts. The most recent example is an AI model developed by Alphabet subsidiary DeepMind that not only helps restore missing text from ancient Greek inscriptions but also suggests when the material was written (within a 30-year period) and its likely geographic origins.
« Inscriptions are particularly important because they are direct sources of evidence… written directly by ancient people themselves, » Thea Sommerschield, a historian and machine learning expert who assisted in the model’s development, told media during a news conference.
These writings are frequently damaged due to their age, making restoration a gratifying challenge. Because they are frequently etched on inorganic materials like as stone or metal, procedures such as radiocarbon dating cannot be utilized to determine when they were written. « To tackle these tasks, epigraphers hunt for textual and contextual analogies in similar inscriptions, » said Sommerschield, who co-led the research with DeepMind staff researcher Yannis Assael. « However, it is extremely difficult for a human to harness all available, relevant data and uncover underlying patterns. »
This is where machine learning can come in handy.
Ithaca is a new piece of software that was trained on a dataset of 78,608 ancient Greek inscriptions, each of which is annotated with metadata describing where and when it was written (to the best of historians’ knowledge). Ithaca, like any machine learning systems, looks for patterns in this data, embedding it in sophisticated mathematical models, and then uses these conclusions to suggest text, date, and origins.
The scientists that developed Ithaca claim that it is 62 percent accurate when recovering letters in damaged texts in a report published in Nature. It can ascribe an inscription’s geographic origins to one of 84 ancient global regions with 71 percent accuracy and can date a document to within 30 years of its known year of writing on average.
These are encouraging figures, but it’s crucial to note that Ithaca cannot function without human assistance. Its recommendations are ultimately reliant on evidence gathered through traditional archaeological methods, and its designers are pitching it as as another tool in a larger set of forensic methodologies, rather than a fully-automated AI historian. « Ithaca was created as a supplement to help historians, » Sommerschield explained.
Ithaca, according to Eleanor Dickey, a professor of classics at the University of Reading who specializes in ancient Greek and Latin sociolinguists, is a « interesting development that may enrich our knowledge of the ancient world. » She did, however, add that a 62 percent accuracy rate for restoring deleted text was not encouraging — « when people rely on it, they will need to keep in mind that it is inaccurate approximately one-third of the time » — and that she was unsure how the program would fit into existing academic approaches.
DeepMind, for example, cited experiments that demonstrated the model improved historians’ accuracy in recovering lost text in ancient inscriptions from 25% to 72%. Dickey, however, points out that those being tested were students, not expert epigraphers. She claims that while AI models are widely available, they cannot or should not replace the small cadre of trained academics who decipher texts.
« It is not yet apparent to what extent usage of this technology by legitimately trained editors will result in an improvement in the editions that are now accessible — but it will be interesting to find out, » Dickey added. She went on to say that she wanted to check out the Ithaca model for herself. The software, as well as its open-source code, is available for anybody to test online.
Ithaca and its predecessor (called Pythia and released in 2019) have already been utilized to aid contemporary archaeological discussions, including assisting in the dating of inscriptions unearthed in Athens’ Acropolis. However, the software’s ultimate potential has yet to be realized.
Sommerschield emphasizes that the true value of Ithaca may be found in its adaptability. Although it was trained on ancient Greek inscriptions, it was easily adaptable to other ancient characters. « The architecture of Ithaca makes it truly relevant to any ancient language, not just Latin, but Mayan, cuneiform; basically any written media — papyri, manuscripts, » she explained. « There are numerous opportunities. »