A Question-Answering Approach to Key Value Pair Extraction from Form-like Document Images