Project Description

VUTT (Early Modern Text Workspace) is a web application for viewing and editing transcriptions of historical documents. The application displays scanned document images and LLM-recognized text side-by-side, allowing for text correction and annotation.

Corpus

The current corpus includes publications from the Academy of Tartu (Academia Gustaviana / Academia Gustavo-Carolina) from the years 1632–1710. The texts originate from the digitized collections of the UT Library and are largely based on copies of publications collected during the compilation of Ene-Lille Jaanson, comp., Druckerei der Universität Dorpat 1632-1710: Geschichte und Bibliographie der Druckschriften (Tartu University Library, 2000).

Recognition Model

The recognition model is page-based. It uses the Qwen3-VL-8B model, which has been fine-tuned on 1500 pages of transcribed text originating from various databases (including inspiration from data at zenodo.org/records/15764161):

• Transcriptions produced at the UT Library research center (conducted within the ETAg research project PUT 132 (“Humanist Greek in Early Modern Estonia and Livonia: A Cultural Bridge to the European Present and Past”; 2013-2016, PI J. Päll), Helleno-Nordica: Humanist Greek in the Swedish Empire (PI Janika Päll, sub-project of the Lund University project “Helleno-Nordica. The Humanist Greek Heritage of the Swedish Empire”, 2016-2022, PI Johanna Akujärvi), and with the support of University of Tartu base-funded grants)
• With the support of the UT Professorship of Intellectual History
• Selected pages from the Gallicorpora dataset (github.com/Gallicorpora/HTR-imprime-17e-siecle)

The texts have been transcribed both manually and using the Transkribus and eScriptorium environments. Transcription authors and reviewers (in alphabetical order):

• Ove Averin
• Meelis Friedenthal
• Jaana Jurtšenkova
• Kristiina Kase
• Vallo Kask
• Kristin Klaus
• Hant Mikit Kolk
• Pärtel Piirimäe
• Agne Pilvisto
• Anni Polding
• Janika Päll
• Kaarina Rein
• Rahel Toomik

Numbers

• 1199 works
• ca 20,000 pages
• Languages: Latin, German, Swedish, Greek, Hebrew

Technology

• Frontend: React + TypeScript + Tailwind CSS
• Search Engine: Meilisearch
• OCR: fine-tuned Qwen3 VL 8B model

VUTT is developed at UT Library. The logo was suggested by Rahel Toomik and comes from the decorative frame of the title page of the sermon Een kort och enfaldigh Lijkpredikan ... Dorpt: J. Vogel, 1642.

Contact

For questions and suggestions, please contact: Meelis Friedenthal

License