Title: Language Generation with Interactions, Explanations, and Commonsense

Speaker: Bodhisattwa Majumder (UCSD)

Abstract: 
Intelligent interactive systems offer great promises as human-machine interfaces for increasing accessibility and user-friendliness. Interactive systems that communicate with humans in natural language are often considered more trustworthy. While recent development in interactive systems (e.g., conversational AI) and language generation has been tremendous, they often struggle to generate responses that are adequately grounded in the real-world context. In particular, these texts often lack commonsense, explanations, and subjectivity -- a long-standing goal of AI.

In this talk, I will partly address these problems in three parts and hint at future possibilities and social impacts. Particularly, we will discuss: 1) methods to capture commonsense implications of input persona in a persona-grounded dialog agent; 2) ways to synthesize human experiences as background stories for engaging dialog generation; and 3) algorithm to generate natural language explanations of machine predictions and investigate the role of commonsense in them.