Reading Bubbe’s Letters Using AI. Unlocking Jewish Historical Documents: AI-Powered Transcription of Yiddish and Beyond

The speakers showcase the power of AI-driven transcription and translation for handwritten Jewish historical documents, focusing on Yiddish as a primary example. The L’Dor V’Dor Foundation’s (LDVDF) Documentation of Jewish Records Worldwide (DoJR) AI Lab develops and fine-tunes AI models to enhance Jewish genealogical research.

The DoJR AI Lab Yiddish Team will share its multi-year effort to train the Transkribus AI tool to convert handwritten Yiddish texts into searchable, machine-readable formats. Training sets include family letters, Joint Distribution Committee (JDC) communications, Lithuanian Jewish community minutes, and Russian Empire vital records.

Through a two-step process—Transkribus transcription and AI-powered translation (e.g., ChatGPT)—we demonstrate practical applications for genealogy, using handwritten letters to the Grajewo Relief Society. We also address challenges in transcribing Yiddish, illustrated by JDC records and 19th-century metrical books from Zaslav, Proskurov (Khmelnytsky), and Kherson.

Attendees will gain insight into the capabilities, limitations, and evolving potential of AI in processing Jewish historical documents across multiple languages.

Presenters

Edna Kalka Grossman was born in the Landsberg, Germany DP Camp to Polish Holocaust survivors. With her surviving grandmother, her family became a three-generation, Yiddish speaking, American household in 1950. She attended a Chabad school where all Jewish subjects were taught in Yiddish. Ultimately, a love of mathematics led to a Ph.D., employment at IBM’s Research Division, and a role as a consultant bringing A.I. projects quickly to bear on IBM customers’ business problems. After retiring, interest in genealogy led her to a volunteer project that required skills rarely found together - fluency in Yiddish and A.I. experience.

Hatte Rubenstein Blejer is a PhD linguist and expert in Natural Language Processing and A.I.  She is the lead for the L’Dor v’Dor DoJR projects to automate transcription from handwritten Yiddish and Ladino. She has studied and/or speaks Hebrew, Arabic, Spanish, Yiddish, Polish, Ladino, French, German, Aramaic, Amharic, Berber, and Romanian. She is a member of the board of LitvakSIG and served on the board of JGSLA as VP of Programs.  She is a Geni curator and a moderator for the Facebook groups Jewish Genealogy Portal and Suwalki-Lomza Independent Jewish Genealogy Research.  

Jan Gronski is L’Dor v’Dor’s DoJR Technology Advisor, building the platform to manage interactions with organizations and individuals, developing partnerships in the community, and discovering potential catalog content. He has considerable hands-on experience in creating A.I. models for automatic handwriting recognition using Transkribus. He is fluent in English, Polish, and Chinese; reads, writes, and speaks Russian, German, and Spanish; and has a basic comprehension of French and Turkish. He is a key volunteer for JRI - Poland. His undergraduate studies in mathematics were done at the University of Warsaw. He has a PhD in mathematics from the University of Illinois.

Click the button to register for this meeting. After registering, you will receive a confirmation email containing information about joining the meeting.