5 février 2019 à 11:30 au LIMSI

Understanding Humans and the Stories they Tell

From ancient times, storytelling has been a strong medium to bring a community together and impart traditions and knowledge across generations. The ability to understand, imagine, and compose stories is a key aspect of human intelligence that differentiates us humans from other species. Inspired by this, my primary research interest is in developing AI that can understand and learn from stories. Complexity in understanding stories is often brought upon by the rich and diverse characters, their relationships, interactions, and emotions presented in different social situations. To learn about them, we analyze and annotate over 50 movies and capture this structured knowledge in the form of graphs. As a direct consequence, the graphs enable us to filter and find video clips with increasing specificity about the situation (e.g., a garden wedding, during the vows, when the husband is talking). The graphs also provide interesting insights about social common sense, and allow us to train neural networks that learn the temporal/causal order in which two characters interact