Скачать или смотреть Helping Wikisource recognize handwritten documents |

The Wikimedia Foundation has partnered with READ-COOP to bring Transkribus, an AI-driven handwriting recognition tool, on Wikisource. This session will introduce Transkribus to the Wikimedia community and share resources around creating new handwriting recognition models to support Wikisource.

The Wikimedia Foundation has partnered with READ-COOP to bring Transkribus on Wikisource in order to support the Wikisource Loves Manuscripts project. The project aims to digitize and transcribe more than 20,000 pages of Indonesian manuscripts to foster their preservation and accessibility.

As the first phase of the project is focused in Indonesia, handwriting models for Indonesian languages are being created with the help of our partner IIIT Hyderabad. Because Google OCR and Tesseract, the two OCR engines already integrated in Wikisource, do not support Balinese and Javanese, Transkribus came into play. Transkribus is AI-powered platform for text recognition, transcription and searching of historical documents. With adequate training material, OCR models can be trained for any language or script and produce automatic transcriptions with a typical accuracy of around 95%.

In this session, members of READ-COOP and WMF will share a demonstration on how Transkribus works and share resources which will help other Wikimedia communities to create new OCR models. We hope that this collaboration will inspire other Wikimedia communities to engage contributors in preserving and transcribing manuscripts as well as unlocking their written past.

Presenters: Satdeep Gill (WMF), Kinneret Gordon, Sara Mansutti