For the thousands of ancient texts discovered, scholars must devote countless hours in deciphering faded writing, filling in missing characters and sections of texts and determining the age and origin of the text.
Restoring Ancient Texts with AI
Ancient inscriptions are vital evidence for understanding past civilisations. Unfortunately, all too often, inscriptions have been damaged to the point of illegibility. To make matters even more complicated, these texts can be found far from their original location, and their date of origin is steeped in uncertainty – meaning historians must attempt to restore the missing text and establish the original date and place of writing.
But now historians have a new tool to help them study ancient inscriptions. “Just as microscopes and telescopes extended the realm of science, AI could support historians, aiding their workflow and improving our collective understanding of ancient written cultures,” says historian and epigrapher Thea Sommerschield.
With the support of the Marie Skłodowska-Curie Actions (MSCA) programme, and in collaboration with Google DeepMind, the University of Oxford and AUEB University in Greece, Sommerschield has developed Ithaca, the first deep neural network aimed at restoring, dating and placing inscriptions from the ancient Greek world.
A collaborative tool for historians
Ithaca’s deep neural network architecture allows it to process large amounts of data and extract textual patterns. Harnessing mutually relevant information, it jointly tackles the tasks of textual restoration and geographical and chronological attribution.
To maximise Ithaca’s value as a research tool, it was designed to produce results that can be easily interpreted by historians while also facilitating collaboration. “We also implemented several visualisation tools to help researchers better interpret Ithaca’s results, providing a richer understanding of ancient history and unlocking the potential for cooperation between AI and historians,” adds Sommerschield.
That potential for collaboration can be seen in the numbers. While Ithaca alone achieves 62 % accuracy when restoring damaged texts, when used by historians it raises their accuracy from 25 % to 72 %. The tool can also attribute inscriptions to their original location with 71 % accuracy, and date them to within 30 years of ground-truth ranges. “This collaborative approach has already yielded results, with key texts of Classical Athens being redated, fuelling new debates in ancient history,” remarks Sommerschield.
Predicting the past with Ithaca
A research paper on the tool was published in the scientific journal ‘Nature’. Furthermore, thanks to a partnership with Google Cloud, a free online interface allows researchers, educators, museum staff and others to use Ithaca for their own research and is receiving hundreds of queries weekly.
Ithaca has also been successfully integrated into school curricula by over 80 teachers across Europe, who are leveraging AI tools to help bridge the gap between computer science and the humanities.
But this is just the start for tools such as Ithaca and the potential for collaboration between AI and the humanities. While ancient Greece plays an instrumental role in our understanding of the Mediterranean world, it’s still only one part of a vast global picture of civilisations.
“Because its architecture makes it easily applicable to any ancient language, from Latin to Akkadian, and to such written mediums as papyri and manuscripts, Ithaca knows no geographic or historic bounds,” says Sommerschield.
The research team is currently training the tool on other ancient languages, but historians can already use their own data sets within the current architecture to study other ancient writing systems.
According to Sommerschield, the MSCA Fellowship provided by PythiaPlus was a pivotal experience, fostering collaboration, expanding networks and paving the way for future breakthroughs. “As we delve deeper into the past with Ithaca, the possibilities for AI in the ancient world are endless,” she concludes.