Is there an AI application known in Germany that can translate a document written by an ancestor in “Old German” (Sütterlin) into “High German”?
I can decipher some parts, but not all.
Try Transkribus and their ‘Super Models’ like German Genius or The Text Titan I. It is an online/cloud service with a user-friendly interface.
If you prefer open-source or local solutions, here are some alternatives:
OCR4all (Local software): Developed by the University of Würzburg. It is a desktop application you install locally (usually via Docker) that uses the Calamari engine for historical scripts.
TrOCR via Hugging Face (Online or Local): Specialized models from the University of Bern. You can test them directly in your web browser on Hugging Face, or run them locally if you have a Python/AI setup.
eScriptorium (Self-hosted/Server): An open-source platform where you can set up your own local server to process documents using the Kraken OCR engine.
UB Mannheim / Tesseract (Local software): The University Library of Mannheim maintains the Windows installers for Tesseract. This is a command-line tool you install on your PC, best suited for printed Fraktur rather than handwriting.
I just spent some time with Transkribus to get familiar with it.
I tried several different settings on my scanner (contrast/brightness).
I have to say my results weren’t as successful as yours, which you posted here.
I also used the “German Giant engine,” but the translation wasn’t much more detailed than what I had already translated myself.
But I’ll keep trying to “tweak” it.
I’ve been using tesseract for a while now (for regular OCR tasks in English).
I’m currently testing Transkribus and will try out the other tools later, then share my experiences.
A test with Google Gemini was not very successful. Completely wrong translation.
I think the result contains too many errors and leads the user astray… The best thing is still to familiarize yourself with the script and learn to read it yourself. Which in this case is really not difficult.
I was able to decipher most of it already. That, along with the translation I received from Transkribus, has brought me a bit further.
The rest will reveal itself in time.
I now know that I need to continue my search in Poland (Goluchow, Schutschen (Szuc), Schutschenofen (Pidun)).
Cheers
Edit: I noticed that the quality of the scanned document (varying contrast/brightness) plays a huge role in how well it can be translated by Transkribus.
I found a list of German/Polish place names on Wikipedia.de that was very helpful to me.
I’m posting the link here so that other interested researchers can find it too.
Hi, I’m currently trying to summarize church records. I have to say Gemini is great at taking something off the bit, and Claude then puts it together perfectly, as a note on that.