@article{fischer-puppe-2026-human,
    title = "Human-in-the-Loop Mass Transcription and Ground Truth Annotation for Challenging Historical Documents",
    author = "Fischer, Norbert  and
      Puppe, Frank",
    editor = "Piperidis, Stelios  and
      Bel, N{\'u}ria  and
      van den Heuvel, Henk  and
      Ide, Nancy  and
      Krek, Simon  and
      Toral, Antonio",
    journal = "International Conference on Language Resources and Evaluation",
    volume = "main",
    month = may,
    year = "2026",
    address = "Palma de Mallorca, Spain",
    publisher = "ELRA Language Resource Association",
    url = "https://preview.aclanthology.org/ingest-lrec/2026.lrec-main.559/",
    pages = "7023--7033",
    abstract = "Challenging historical documents still pose significant difficulties for fully automatic layout detection and text recognition, requiring lengthy, demanding correction. We describe our experiences with complex layouts and present our workflow with AdaptOCR, a web-based annotation tool designed to facilitate the efficient transcription and ground-truth annotation of demanding historical documents. Addressing the limitations of existing solutions, AdaptOCR prioritizes a streamlined workflow with an integrated trainable layout and OCR pipeline. The tool uses the PAGE standard to represent document structure and enables the annotation of baselines, regions, text lines and the correction of their transcriptions providing automatic OCR invocation and dictionary-based error detection. Furthermore, it supports flexible annotations with custom element types and attributes to cater to different project requirements. We demonstrate the effectiveness of the workflow and tool in two demanding applications: The transcription of a large corpus of historical printings and the detection / annotation of handwritten artifacts within the private library of the Grimm brothers. In addition, we evaluate the dictionary-based correction and assess the efficiency improvements using AdaptOCR in a pilot study."
}