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.
Normally, epigraphers – scholars who study ancient texts written on durable materials such as stone, metal or pottery – use their own knowledge of repositories of information and digital databases performing “string matching” searches to find textual and contextual parallels.
However, differences in the digital-search query that can exclude or conceal relevant results and that the texts often are discovered not in their original context can complicate the work.
In a new research paper published in the peer-reviewed scientific journal Nature, Thea Sommerschield, a historian and Marie Curie fellow at Ca’ Foscari University of Venice and a fellow at Harvard University’s Center for Hellenic Studies, and Yannis Assael, a research scientist at Google’s artificial-intelligence research lab DeepMind, presented their Ithaca state-of-the-art AI technology.
Ithaca is the first deep-neural network meant to improve the task of restoring and attributing ancient texts.
It uses deep learning for geographical and chronological attribution for the very first time and on an unprecedented scale, the researchers said.
An AI deep-learning model is based on a neural network inspired by the biological neural networks we have in our brains, which can discover and utilize complicated statistical patterns in large amounts of data.
Recent increases in computational power have enabled these models to tackle challenges of growing sophistication in many fields, including the study of ancient languages, the researchers said.
Ithaca is trained on the largest dataset of ancient Greek inscriptions and across the ancient Mediterranean world between the seventh century BCE and the fifth century CE, the researchers said.
These texts were chosen because of the varied contents and context already existing for them and because of the available digitized database for ancient Greek, they said.
Ithaca is meant to be a collaborative technology, focusing on collaboration, decision support and interpretability, the researchers said.
While Ithaca alone achieves 62% accuracy when restoring damaged texts, combining it with the use by historians improved the results’ accuracy from 25% to 72%, they said.
Ithaca includes different visualization aids to increase the ability of researchers to interpret the results, allowing them to evaluate multiple possibilities.
This could help historians increase precision by narrowing the wide or vague date brackets they must sometimes resort to, the researchers said.
They added that their research shows how models such as Ithaca can “unlock the cooperative potential between artificial intelligence and historians,” impacting the way one of the most important periods in human history is studied and written about,
