Assess the value of severity-weighted OCR evaluation for downstream tasks

Determine the value of the severity-weighted character error rate for downstream tasks.

Background

The paper motivates unequal OCR error costs by arguing that some transcription errors are more harmful to readers than others, and it validates the proposed severity measure against reader judgments and preferences. It does not, however, evaluate whether the metric predicts or improves performance in downstream applications such as retrieval, question answering, or other document-processing tasks. The authors explicitly defer this unresolved assessment to future work.