The DOCX extractor handles tables by embedding them as JSON within text chunks instead of separate DuckDB tables.