Decoding the Past: Exploring AI-Based Handwritten Text Recognition in Digital Collections
At a university library, the Digital Services Department recently completed a pilot project evaluating the potential of AI-powered Handwritten Text Recognition (HTR) tools for use with historical materials. The project tested seven products to determine which systems produced the most accurate transcriptions for a specific item. Conducted in collaboration with a student researcher, this initiative provided valuable insights into workflow design, data management, and quality control while laying the groundwork for future AI evaluation projects. This presentation will share key findings from the HTR testing process, including comparisons of model performance, challenges with historical handwriting, and strategies for improving transcription accuracy. Attendees will learn how small-scale testing can inform scalable, ethical, and sustainable AI integration within library digitization programs.