Completely agree that dialogue management is a complex and interesting field of study in NLU/NLP. I’m intrigued by the level of complexity the BOTS will deal with understanding the emotion, sarcasm, optimism or lack thereoff and few other shuttle side of human conversation. My thoughts on that, we can solve two ways, and they can go hand in hand. One, more deep learning via sniffing and processing real-world conversations. Second, systematically remove such shuttle side from human conversation, making it plain and simple for machine understanding. Humanly conversation will be thing of past soon, or only available in literature.