Measuring Search Retrieval Accuracy of Uncorrected OCR: Findings from the Harvard-Radcliffe Online Historical Reference Shelf Digitization Project

This report presents the findings of an investigation to evaluate the conditions for search retrieval successes and failures when using uncorrected OCR for indexing. The purpose of the study was to assess whether low-cost, high-production techniques for text conversion were adequate to produce digital reproductions of consistent quality and usability. We sought to identify attributes of the original material or the OCR-produced text that could predict when additional, costly processes (OCR correction or keying) would be needed to meet retrieval requirements for text digitization projects.