One of the most notorious debates of recent decades within the academic community studying the epigraphy of ancient Greece focuses on public documents issued in Athens in the mid-5th century BC. Its interpretation is considered essential to understanding Athenian imperial politics at the time and one of the most transcendental periods in classical history. Traditionally, these inscriptions have been dated to around 446/5 BC. J.-C. according to the way of writing the letter sigma, hitherto composed of three bars (ϟ) and not of four (Σ). However, several historians have suggested a somewhat later chronology.
A new artificial intelligence tool called Ithaca in honor of the mythical Hellenic island, homeland of the hero Ulises, and developed by an international team of researchers was finally able to resolve this dispute. For example, the famous decree of Chalcis, kept in the Acropolis Museum, which contains an oath of loyalty made by the city of the island of Euboea to Athens, has been dated to 420 BC. while the decree of Kleinias, which regulated the collection of tributes throughout the Athenian Empire, has been dated to a very similar date, 421 BC. These discoveries delay the publication of the texts by more than twenty years.
Ithaca is a deep neural network developed to simultaneously perform the tasks of restoration of texts of ancient Greece – originally inscribed on stone and in many cases fragmented and incomplete – and geographical and chronological cartography. The system was built around a large database of almost 200,000 engraved inscriptions in the Greek language throughout the Mediterranean between the 6th century BC and the 5th century AD.
As detailed in an article published in Nature the tool’s developers, led by Yannis Assailfrom the British company DeepMind, and Thea Sommerschield, from the University of Venice, “Ithaca can uncover epigraphic motifs on an unprecedented scale and in unprecedented detail.” For the reconstruction of fragmentary texts, the machine offers researchers the twenty most probable predictions. As for chronological dating, the program divides all dates between 800 BC. AD and 800 AD. map and a bar graph.
The main conclusion of the experiment carried out by the researchers ensures that Ithaca’s accuracy is 62% when it comes to restoring damaged texts independently, a figure that rises to 72% when manipulated by a historian specialized in the epigraphy of ancient Greece. Its success in locating the place of origin of the inscriptions has been dated one point less.
Correct the mistakes
The potential of the tool seems indisputable. It is not a question of displacing the conscientious work of academics, but of strengthening it so that one can constantly reassess the knowledge of history, according to Charlotte Rouche, from the Classics Department of King’s College London. A new word or phrase can provide relevant data to shed light on questions of domestic life, politics or economics.
“One of the priorities of our interdisciplinary team was to make the Ithaca results interpretable by historians: instead of providing a single result, we propose various prediction hypotheses and we visualize model certainty in a distribution,” Assael told SINC. “At the same time, we present which words contributed the most to a specific prediction. These visualizations allow experts to use their contextual information to choose the production or a more appropriate result, thus being able to shed light on unexplored historical knowledge”
Many inscriptions that have survived to this day, a source of basic information for the study of past civilizations, have been lost in whole or in part over the centuries. Besides the obvious illegibility of the incomplete documents, it is impossible to date many of these texts because they do not contain any organic material that would make radiocarbon analysis possible. The aforementioned decrees from Athens from the middle of the 5th century BC are a good example of what can be achieved with this algorithm.
“The fact that Ithaca was formed on the largest available data set of Greek epigraphic texts [el corpus del Packard Humanities Institute de Santa Clarita, en California, o el Lexicon of Greek Personal Names de la Universidad de Oxford] makes possible challenge or overcome individual biases or, indeed, errors in the existing scientific tradition, despite the fact that the dataset in question is originally based on scientific accumulation.
This program focused on ancient Greek texts due to the wide variety of content and contexts of the epigraphic record and access to a digitized corpus. “Ithaka is the first model of epigraphic attribution and restoration of its kind”, conclude the researchers, while emphasizing its usefulness for other disciplines that study past writing, such as papyrology or numismatics. The most similar project is Pythia, made in 2019 by the same team, another algorithm for reconstructing the remains of classical epigraphy. Artificial intelligence challenges our knowledge of history.