Advancing Handwritten Text Recognition for Non-Latin Scripts (NonLatinHTR) aims to develop two open-access HTR models for Greek and Arabic scripts. While Latin-script manuscripts and printed texts have benefited from extensive research and recognition tools, other Mediterranean scripts remain significantly underserved. This disparity stems from a scarcity of open-access digital collections, methodological challenges in adapting Latin-based paleographical frameworks to other scripts, and the technical expertise required in statistics and AI.

Although deep learning techniques, such as neural networks, have proven highly effective for transcription, this project aims to go further by using pattern recognition to assist paleographers in their research—particularly through the analysis of large datasets. Can machine learning achieve the same analytical conclusions as traditional paleography? Can it uncover subtle patterns unnoticed by human experts? How can these computational tools enhance our understanding of scribal practices and workshop traditions? This project will address and test these key questions during a research stay at the VMA Lab (CSMC).