Artificial intelligence could bring to life lost texts, from imperial decrees to the poems of Sappho, researchers have revealed, after developing a system that can fill in the gaps in ancient Greek inscriptions and pinpoint when and where they are from.
Dr Thea Sommerschield, a co-author of the research at Ca’ Foscari University of Venice and Harvard University, said inscriptions were important as they were written directly by ancient people and were evidence of the thought, language, society and history of past civilisations.
“But most surviving inscriptions have been damaged over the centuries. So their texts are now fragmentary or illegible,” she said, adding that they may also have been moved from their original location, while methods such as radiocarbon dating were unusable on materials such as stone.
Writing in the journal Nature, Sommerschield and colleagues report how they built an AI system that they nicknamed Ithaca, after the Greek island that was home to the legendary King Odysseus.
The team fed Ithaca more than 63,000 transcribed ancient Greek inscriptions, enabling it to pick out patterns in the order of letters and words, as well as associations between words and phrases and the age and provenance of the text.
The team then tuned the system before exploring whether it could accurately suggest when and where another 7,811 inscriptions were from, and propose a selection of letters and words to fill in artificially created gaps in the inscriptions, ranked by probability.
The results reveal that Ithaca achieved 62% accuracy when used alone to fill in the gaps in inscriptions, and 72% accuracy when the system’s suggestions were interpreted by a historian – about three times higher than when historians worked alone. The team said Ithaca was able to date the inscriptions to within 30 years of their established date and correctly identified their provenance 71% of the time.
“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,” said Dr Yannis Assael, a co-author of the work from the AI company DeepMind.
The team said the approach could be used for any medium and any ancient written language, from Latin to Cuneiform, and it might be possible to train the system on Greek literary texts written on fragments of papyrus – an approach that could shed light on the writings of poets such as Sappho. There is also the potential to develop AI systems that could provide insights into the authorship of texts.
The researchers said Ithaca had already been used on a set of decrees most of which were found on the Acropolis of Athens,suggesting one – relating to the collection of tributes across the Athenian empire – dated to 424BC rather than 448-7BC as was long thought, chiming with recent dating breakthroughs.
“Although it might seem like a small difference, this 30-year shift has momentous repercussions for our understanding of the political history of classical Athens, and helps us better align literary sources – such as Thucydides’ account of these years and events– with the epigraphic record,” said Sommerschield.
Prof Peter Liddel, an expert in Greek history and epigraphy at the University of Manchester who was not involved in the research, said even the provenance of many of the marbles brought back by Lord Elgin was unclear.
“The application of AI through Ithaca certainly has the potential to contribute to the toolbox of historians involved in analysing ancient texts and using them to understand processes like the development of imperialism or the nature of cult activity,” he said.
However, Liddel warned that, like scholars, AI was limited by gaps in the ancient record. “AI is only powerful as a tool to help us ask questions about, and make comparisons to, the existing evidence,” he said.
Prof Melissa Terras, an expert in digital cultural heritage at the University of Edinburgh, said it was important to keep training scholars in traditional approaches to be able to develop AI systems such as Ithaca, and to interpret the suggestions they generate. But she said there was huge potential for AI to assist with interpretation of the past and its cultures given ancient texts were often fragmented yet followed structured formats.
“This means they require a lot of cross-referencing for the human brain to solve the puzzle – but this is the type of repetitive calculation that [AI systems such as] deep neural nets excel at,” she said.
