People-centric Natural Language Processing

Machine learning and natural language processing provide a framework for extracting meaning from text, and have given us great advances over the past fifty years in areas as diverse as machine translation, question answering, and information retrieval. Many of the written texts that we apply these techniques to â news articles, emails, social media, books â are the product of a profoundly social phenomenon with people at its core. People are the authors of text, they are its audience, and often the subject of that text itself: news articles detail the roles of actors in current events, social media (including Twitter and Facebook) documents the actions and attitudes of friends, and books chronicle the stories of fictional characters and real people alike.

In this talk, I will present a set of probabilistic models that learn patterns of identity and behavior in descriptions of people in text. Unsupervised models of biographical structure allow us to learn the way different life events (such as graduating high school, marriage, and becoming a citizen) are described in text, along with the typical times in a person's life when they occur; and unsupervised models of personas allow us to learn abstract entity types from the stereotypical actions they perform. This work reveals large-scale patterns in descriptions of people while also uncovering implicit biases of the authors; I argue that developing computational models that capture the complexity of the interaction of people with text will yield deeper, socio-culturally relevant descriptions of these actors, and that these deeper representations open the door to socially-aware language technologies that have a more useful understanding of the world.

Speaker: David Bamman