Vilne Yiddish Handwritten Dataset (HTR, 205 pages)

Version 1.0, 2025-11-13, Sergii Gurbych ([e-mail])

This dataset for Handwritten Text Recognition includes layout segmentation (TextRegions, TextLines, and line polygons) and Unicode transcriptions in PAGE XML for 205 images of handwritten Yiddish texts. The materials originate from autobiographical, epistolary, and historical manuscripts that were created between 1930 and 1940 and are stored in the following institutions: the YIVO Institute for Jewish Research (New York, published at digipres.cjh.org), the Judaica Research Center of the Martynas Mažvydas National Library of Lithuania, and the Central State Historical Archives of Ukraine (Lviv, access to the documents through the digital copies held in the Library of the University of Illinois in Urbana-Champaign).

The dataset was prepared using the open-source eScriptorium platform and Kraken HTR engine. Transcriptions are line-based and manually corrected. The dataset contains 4411 lines and approximately 45,600 words. Each PAGE XML file is accompanied by its corresponding image file in JPG format.

Images included here are publicly accessible via their respective institutions. No explicit reuse restrictions were stated as of November 13, 2025. Their use here is subject to any institutional conditions that may apply. More detailed information about each document is provided in the README.txt file.

This project was funded by the European Union through the Economic Recovery and Resilience Facility NextGenerationEU, under Lithuania’s national plan “New Generation Lithuania,” the funding body is the Central Project Management Agency (CPMA), project CAPADHY, project number: 10-036-T-0005. The project was carried out at the Faculty of History, Vilnius University.

This dataset is free to use for non-commercial purposes under CC-BY-NC-SA 4.0.