Vision is Not Enough: Enhancing Bounding Box Grounding for Document Key Information Extraction with OCR Anchors

Examine how text detected by OCR helps an extraction model point to the correct evidence on a document page.