A Hybrid Handwritten Text Recognition Workflow: Evaluating Transkribus and eScriptorium in the Joseph Hooker Correspondence Project.
JHCP-updated-report_15.07.2025.pdf
799.27 KB
Adobe PDF
33f9b34131df2ad6969c56fbace91a8f
This report presents a hybrid Handwritten Text Recognition (HTR) workflow developed through the Joseph Hooker Correspondence Project (JHCP) at the Royal Botanic Gardens, Kew. Combining Transkribus as a volunteer-facing transcription interface with eScriptorium for open-source model training, the study demonstrated that once a sufficient ground truth dataset was established, error rates between the two platforms became marginal. Volunteer surveys and interviews illuminated the practical challenges of implementing HTR in resource-limited, volunteer-dependent projects. The resulting model and training data were published openly via Zenodo and HTR-United, and the report concludes that hybrid Transkribus/eScriptorium workflows offer a cost-effective, reproducible framework for small-scale archival transcription — supporting broader calls for open-source model sharing across the cultural heritage sector.