Conversational systems are becoming an integral part of our daily routines. Every day, millions of people use natural-language interfaces via in-home devices, phones, or messaging channels such as Messenger. We strive to create new conversational technologies that have a deep understanding of the conversation and the context around it and deliver a personalized experience to the user that is both task oriented and empathic.
The next generation of conversational AI systems will be multi-modal and pro-active, integrating cues across several modalities to provide creative and on-spot response to the users.
In this work, we examine two controllable neural text generation methods, conditional training and weighted decoding, in order to control four important attributes for chitchat dialogue: repetition, specificity, response-relatedness and question-asking.