A New Framework for Error Analysis in Computational Paleographic Dating of Greek Papyri

TL;DR

This paper introduces Hell-Date, a precisely dated dataset of Greek papyri from the Hellenistic period, and a CNN-based dating framework (fCNN) to predict exact production years from line images. It emphasizes an error-aware evaluation that accounts for the inherent imprecision of palaeographic dating through an Error Time Window (ETW) and associated metrics, showing that the AI model achieves performance comparable to human experts. The study systematically compares AI dating to human dating, analyzes transfer learning between Hell-Date and a Roman-period PLL dataset, and demonstrates that simple dataset fusion does not guarantee better results. The findings suggest that contemporary computational dating can match expert performance on Greek papyri and establish a solid foundation for scaling up datasets and refining handwriting-style analysis. The proposed framework offers a principled way to quantify dating accuracy in the presence of temporal imprecision, informing future developments in digital palaeography and computational dating.

Abstract

The study of Greek papyri from ancient Egypt is fundamental for understanding Graeco-Roman Antiquity, offering insights into various aspects of ancient culture and textual production. Palaeography, traditionally used for dating these manuscripts, relies on identifying chronologically relevant features in handwriting styles yet lacks a unified methodology, resulting in subjective interpretations and inconsistencies among experts. Recent advances in digital palaeography, which leverage artificial intelligence (AI) algorithms, have introduced new avenues for dating ancient documents. This paper presents a comparative analysis between an AI-based computational dating model and human expert palaeographers, using a novel dataset named Hell-Date comprising securely fine-grained dated Greek papyri from the Hellenistic period. The methodology involves training a convolutional neural network on visual inputs from Hell-Date to predict precise dates of papyri. In addition, experts provide palaeographic dating for comparison. To compare, we developed a new framework for error analysis that reflects the inherent imprecision of the palaeographic dating method. The results indicate that the computational model achieves performance comparable to that of human experts. These elements will help assess on a more solid basis future developments of computational algorithms to date Greek papyri.

Figures (5)

• Figure 1: Image of the papyrus TM 244 $=$ P.Lond. III 1208 $=$ P.Lond. inv. 1208. Image courtesy of the British Library. The first 3 lines of column 2 translate: "When Ptolemy, also known as Alexander, and Berenice, his sister, the mother-loving gods, are reigning for the 17th year, under the priests and priestesses and the canephoros which are in charge, day 4 of the month Mecheir, in Pathyris, under the notary Ammonios".
• Figure 2: Accuracy of the Hell-Date test set dating according to the size $\alpha$ of the ETW.
• Figure 3: Datings by fCNN (on the left) and the experts (on the right) a box plots per document. The ground truth (actual date of writing) is indicated with an X.
• Figure 4: Comparison between human experts and AI Model performance.
• Figure 5: Comparison between computational model performance (left) and best-performing Human Expert (right).