Beyond VQA: Generating Multi-word Answer and Rationale to Visual Questions