First, our tables model now works better with complex table structures that contain split cells and merged cells, which make it difficult to align cell values to the correct column header or row header. Next, Amazon Textract has further improved in identifying rows and columns for cells with wrapped text (text present across multiple lines), even for tables without explicit boundaries. Amazon Textract now does a more accurate determination of cells with content across multiple lines as opposed to when it is a new row without an explicit boundary. Finally, Amazon Textract has also improved the forms model to give more accurate results for key-value pair identification. These benefits apply to many types of documents, but are especially pronounced for documents where tables and key-value pair are present within the same page. Now, Amazon Textract correctly identifies key-value pairs embedded within a table.
You can learn more about these updates here.