Our research project is part of the Visual Media Lab, headed by Professor Jihad El-Sana, the
Department of Computer Science at Ben-Gurion University of the Negev, Israel. This is a
project in Digital Humanities that brings together researchers from both fields, Computer
Science and Humanities. The data we work with is incorporated from many different libraries
through an effort of 60 years to establish and define the principles of Hebrew Paleography and
Hebrew Codicology. This vast dataset, namely SfarData, includes the full codicological
descriptions and paleographical definitions of all dated medieval Hebrew manuscripts (the end
date is 1540). The other major source of our data is the Institute of Microfilmed Hebrew
Manuscripts at the National Library that has undertaken to collect copies of all extant Hebrew
manuscripts from all over the world. As of now, this collection grants access to 76,000
manuscripts in the form of digital images or microfilms.

Currently, manuscripts are classified manually by specially trained paleographers and
codicologists. This is a laborious and lengthy process, which is almost impossible to apply to
large quantities of manuscripts. Also, even the most basic cataloging of a manuscript collection
requires a specially trained researcher in situ and in his absence the collection remains
inaccessible.

Classification of Hebrew scripts changes over time as Hebrew paleography and
codicology evolve. We are using the current version used at the SfarData that divides Hebrew
script prior to 1540 into fifteen sub-types. Each of these sub-types has its own unique
paleographic characteristics that distinguish it from others. These paleographical and some of
the codicological features can be used in machine learning.

Our project aims at developing two-level deep learning models for fine-grained writing
style classification: inter-type and intra-type. The inter-type models classify documents into
regional, graphical, and refined sub-types, according to the study of Beit-Arié et al. The intratype
models cluster the various handwriting documents within a script type into finer clusters;
e.g., according to the handwriting of different scribes. The fine-tuned algorithm will be
incorporated into a web-based system, which includes a user friendly front-end and the deep
learning algorithms on the back-end. Users will upload manuscript samples and obtain their
paleography analysis automatically. As the prior stage of the project, we have trained a deep
learning model to distinguish Ashkenazi semi-square script among other Hebrew scripts, and
obtained a 98% accuracy rate. Currently, we are building a deep-learning model for classifying
Hebrew document into one of X different classes, where X is the number of script types (e.g.,
Ashkenazi, Sephardic, Italian, Oriental, Byzantine, Yemenite, and their sub-types).

In our paper, we intend to talk about the progress of our research, about the manuscripts
and tools that we use, and how we apply deep learning methods to aid paleography analysis of
historical documents.

Deep learning for paleographic analysis of medieval Hebrew manuscripts

The international conference #DHJewish - Jewish Studies in the Digital Age will bring together more than 60 scholars and heritage practitioners to discuss how the digital turn affects the field of Jewish Studies. The conference is organised by the [Luxembourg Centre for Contemporary and Digital History (C²DH).](https://www.c2dh.uni.lu/) 

[For more information, and a PDF download of the program, please see also here.](https://www.c2dh.uni.lu/events/dhjewish-jewish-studies-digital-age) 



In order to meet other participants and for informal chats in the breaks, join us in our [wonder.me room](https://www.wonder.me/r?id=73a12756-8d93-4871-9374-143253bfbf59).