RESEARCHInvestigateNEXT 12 MONTHS
When Low CER is Not Enough: An Analysis of Hallucinations in Vision-Language OCR Systems on Historical Uruguayan Documents
arXiv cs.LG — Machine Learning
Factual evidence
What the source reports
Research finds Vision-Language Models (VLMs) for OCR hallucinate on historical documents despite low character error rates (CER), challenging their suitability for archival transcription.
OneBench interpretation
Institutional assessment
So what
This study highlights a critical gap between benchmark performance and real-world reliability for VLMs in document intelligence, specifically around hallucinations in OCR.
Do what
Your due diligence for VLM-based document processing must extend beyond standard accuracy metrics to include hallucination risk, especially for unstructured historical data sources.