Media release

From: Springer Nature

AI restores and contextualizes Latin inscriptions

An artificial intelligence (AI)-based tool that can predict missing parts of Latin inscriptions from the Roman period is reported in Nature. The tool, named Aeneas, can relate the texts to others, offering contextualization for historians.

It is estimated that around 1,500 Latin inscriptions are discovered every year, which may provide insights into the cultural and linguistic life of the Roman empire. However, words and sentences are sometimes lost over time. Restoring these texts and placing them in the correct geographies and timelines requires historians to place them within the wider linguistic and historical setting by identifying parallels with other texts. This contextualization is often time consuming and highly specialized, requiring extensive knowledge of individual periods.

Yannis Assael, Thea Sommerschield and colleagues present Aeneas, a generative neural network that is able to contextualize ancient texts. The tool can predict missing text, even when the length is uncertain, and is able to suggest context and text parallels. Aeneas has also been developed to include visual imagery in its considerations. To evaluate the network’s potential, the authors launched a collaborative study with 23 historians using the model in a real-world research scenario to assess inscriptions dating from the seventh century bce to the eighth century ce. The historians reported that the context suggestions provided by Aeneas were useful in 90% of cases and improved their confidence in key tasks by 44%. When historians were paired with Aeneas, there were better results in restoration and geographical attribution tasks than with the historians or AI alone. Aeneas was also able to provide dating within a 13-year time frame.

The authors suggest that Aeneas represents a transformative tool that can assist historians and expand our understanding of the past.