Friday, September 4, 2026

Anagnostes: The first AI-driven system that automatically transcribes Greek papyri from images

Greek papyri are a unique and remarkable group of manuscripts. Written mainly between 2,300 and 1,300 years ago, they offer a rare firsthand glimpse into the ancient world and the early Middle Ages. Preserving everything from everyday documents to literary and religious texts, they provide invaluable evidence for historians, philologists, theologians, and other scholars seeking to understand the people, cultures, and ideas of the past.

Papyrus was the primary writing material in antiquity before the widespread adoption of parchment and paper. As a result, texts written on papyrus shed light on virtually every aspect of ancient life. To name just a few examples, papyrus documents constitute the largest body of administrative, economic, and legal texts surviving from the Roman Empire. Private letters and family archives preserved on papyrus provide evidence about social groups often underrepresented in historical sources, including women and children. Thanks to papyri, we now possess literary works by ancient poets and philosophers that were long thought to have been lost during the Middle Ages. Papyrus texts also include the earliest surviving copies of the Gospels and other works of early Christian literature.

Most papyri were—and still are—found in Egypt, where they survived thanks to the dry climate. After major discoveries of large numbers of papyri in the late 19th century, papyrology was born—the discipline devoted to deciphering, reconstructing, analysing, and understanding fragments of papyri. The main task of every papyrologist is to read and interpret a papyrus correctly and, in doing so, to bring back to life a text that has remained unread and silent for hundreds of years. Since the beginning of the 20th century, several generations of papyrologists have prepared scholarly editions of tens of thousands of new texts, thanks to which our knowledge of the past has grown and deepened immensely.

Papyrology as a discipline has always been open to new technologies and is proud to be one of the most technologically advanced fields in the study of the ancient world. Since the recent progress in machine learning, there has been a great deal of interest in how it can be applied to the papyrological work. The question of whether AI can be trained to transcribe the text on papyrus has been constantly raised, sometimes with concern, in other cases with excitement, and always with a degree of scepticism. Today, we can answer this question with a confident "yes".

We present Anagnostes—the first AI-driven system that automatically transcribes Greek papyri from images, developed by Elena Chepel and Anton Repushko.

The model

Anagnostes is a Handwritten Text Recognition (HTR) system powered by advanced machine-learning methods. It has been trained on both literary and documentary papyri dating from the 3rd century BCE to the 7th century CE and currently has an average character error rate (letters CER) of 17.9 % across all types of handwriting, with the best results reaching as low as 1.9–8.6%. It can read not only well-preserved, neat hands but also cursive handwriting and damaged fragments.

 


 

No comments:

Post a Comment