Learning End-to-End Goal-Oriented Dialog with Multiple Answers