Hand-written text recognition software can turn manuscripts into searchable transcriptions. Learn how to use these tools and what they can (and cannot) do for you. The presentation includes a demonstration of Transkribus.
As part of the exclusive “Artificial Intelligence for Genealogists” series, this webinar explores how AI-powered tools—specifically the software Transkribus—are revolutionizing the transcription of historical handwritten records. For genealogists, this technology represents a transformative opportunity to uncover hidden ancestral details, make vast record sets searchable, and overcome longstanding research brick walls with increased accuracy and efficiency.
Transkribus Offers Powerful AI-Driven Transcription for Handwritten Records:
Transkribus uses advanced neural networks trained on historical documents to convert scanned handwritten texts into searchable machine-readable formats. This makes it especially useful for processing large, unindexed archives, even in unfamiliar languages or difficult handwriting styles.
Handwritten Text Recognition (HTR) Outperforms Large Language Models (LLMs) for Transcription Accuracy:
Hoitink’s comparison between Transkribus and tools like ChatGPT revealed that purpose-built HTR systems provide significantly more reliable and text-faithful transcriptions, especially with obscure names or foreign scripts. Unlike LLMs, Transkribus does not “hallucinate” plausible but incorrect content.
HTR Enables Advanced Research Strategies and Breakthrough Discoveries:
By transcribing thousands of pages from Dutch court records, Hoitink uncovered over 100 relevant documents for a single client, reconstructing family lines back to the early 1500s. She also shared methods for using HTR outputs in full-text searches—such as wildcard searches, associative name targeting, and land description matching—to find records beyond traditional indexing systems.
To see how this powerful technology can elevate your own research, be sure to watch the full webinar. Yvette Hoitink’s hands-on demonstrations and real-world case studies provide practical guidance and a clear understanding of both the strengths and limitations of AI-driven transcription tools.
For those ready to dive deeper, don’t miss the wealth of supplemental materials provided in the webinar syllabus. It includes step-by-step instructions, model selection advice, and links to further tutorials—essential tools for any genealogist looking to expand their digital skillset and uncover new ancestral insights through the power of AI.
Facebook
X
Syllabus
Chat Log
I used ChatGPT for transcribing a deed and it totally hallucinated. It was a good reminder to check everything after a transcription.
Sorr, I meant 18th century Spanish.
Can this app transcribe from 18th Spanish?
There are several models for Spanish that are trained on a variety of records, which I think should be able to handle 18th century. The free model Coloso Español is trained on texts from the medieval to the modern period, so that’s where I would start.
Fantastic presentation with very thorough examples and explanations. Loved the side-by-side example of a transcription between Transkribus and ChatGPT. I myself have had the experience of the LLM completely hallucinating the results of the document. I feel prepared to give Transkribus a try! Thank you!
There was a question about which standard model to use for English, that I promised to answer in the comments. I would recommend the English Eagle. But do play around with different models, since they were all trained on different data and your document may resemble one model’s training data better than others.
Fantastic presentation!
Wonderfully thorough with lots of advice. I have 20 years of letters from my mom to scan and transcribe. Sounds like a great option.
Great webinar! Very clear introduction to how to use Transkribus and a great overview of the advantages and benefits of using it. This webinar has inspired me to try it.