can you explain more specific in Document Understanding that i can use to train extractor as we can train classifier.