Austrian Academy of Science releases Apollo Ancient Greek language model

Austrian Academy Releases Apollo Ancient Greek Language Model

The Austrian Academy of Science announced on Tuesday the release of Apollo, the world's first advanced large language model for Ancient Greek, developed with Mistral and Sail Reply.

The model, called Apollo, is trained on roughly 600 million historical Greek words drawn from ancient manuscripts, papyri, and inscriptions. It will be freely available to academics through a chatbot interface to help them identify papyrus fragments for their research.

Where ancient documents are tattered and torn, Apollo is built to quickly fill in the blanks with the most statistically likely words or passages for scholars.

Apollo Adapts to Context and Preserves Human Expertise

Anna Dolganov, a historian and papyrologist at the Austrian Academy of Science, says Apollo adapts to the context of the text. When it sees Homer, it supplements Homeric Greek, and when it sees an inscription in Doric dialect, it uses Doric.

Dimitris Vlitas, partner at Sail Reply, tells WIRED that unlocking new knowledge in this way was truly unthinkable a single year ago for ancient Greek studies.

Apollo is built to propose a selection of word options for a scholar to select between, heading off concerns about errors. Dolganov says the crucial point is that human competence needs to always remain very central to the entire overall process.

Experts Debate Apollo's Impact on Ancient Greek Studies

Stephen Colvin, a professor of classics and historical linguistics at University College London, says very few people in the world are that good at Greek history.

Colvin says Apollo is really very unlikely to change our broad basic understanding of the entire ancient world at all. He says new plays by Sophocles are not going to ever happen because the model cannot create truly new literature.

Armand D'Angour, a professor of classical languages and literature at the University of Oxford, says a machine telling him possible words would speed up matters considerably.

NIST Algorithm and Vesuvius Challenge Decipher Herculaneum Scrolls

A separate effort, the Vesuvius Challenge, combines digital unwrapping with crowdsourced machine learning to decipher badly charred ancient Herculaneum scrolls. The scrolls, buried by Mount Vesuvius in 79 C.E., are too fragile to be physically unrolled without completely destroying them.

A NIST team newly modified researcher Cyrus Daugherty's algorithm originally developed to virtually unroll lithium-ion battery layers to just perhaps help decipher the ancient carbonized papyri.

The team reports in PLOS One that a common X-ray spectrometer can rapidly and nondestructively identify lead in a material. They also propose a new process for deciphering the ancient Herculaneum scrolls using the spectrometer and their team's successful algorithm.

USF Team Uses AI to Excavate Ancient Roman Villa

A University of South Florida team successfully used artificial intelligence this summer to greatly speed up the excavation of a large ancient Roman villa in Sicily.

Davide Tanasi, a professor of digital humanities at USF and the director of IDEx, said the fieldwork was a resounding success. He said the team utilized artificial intelligence to greatly work more efficiently without always replacing any human creativity or decision-making.

Tanasi said AI greatly helped them identify an ancient statue fragment and suggest the likely pattern of a mosaic floor, thereby guiding their important next searches.