Zero-Shot Rationalization by Multi-Task Transfer Learning from Question Answering - View it on GitHub