Closing the parsing gap: reaching SOTA RTL parsing by leveraging LTR capabilities

Current document parsing strategies perform poorly on right-to-left (RTL) languages like Hebrew and Arabic, creating significant issues for RAG architectures. A novel technique called Word Shape Encoding was developed to address this parsing gap by translating RTL documents into an easy-to-parse left-to-right (LTR) format. This method works by mapping each RTL word to an English word with a similar visual geometry and bounding box, which successfully preserves the document's original layout integrity. The approach significantly improves parsing quality for RTL documents, demonstrating that graphical structure is more important than semantic content for accurate parsing.