log in  |  register  |  feedback?  |  help
Knowledge Mismatch in Communicative Interaction: Probabilistic Weighing of Perspectives
2124 H. J. Patterson Hall (Language Science Center)
Wednesday, April 19, 2017, 11:00 am-12:00 pm Calendar
  • You are subscribed to this talk through .
  • You are watching this talk through .
  • You are subscribed to this talk. (unsubscribe, watch)
  • You are watching this talk. (unwatch, subscribe)
  • You are not subscribed to this talk. (watch, subscribe)

Knowledge mismatch is inherent to communicative interaction, since conversational partners naturally differ in their knowledge and beliefs.  Effective conversational moves, such as asking a question or referring to an entity, crucially depend then on each participant acting appropriately with respect to which knowledge is privileged (unique to that participant) and which is shared among the participants.  For example, much work in linguistics suggests that a definite referring expression such as the knife must be interpreted in the context of mutual knowledge (“common ground”) in which there is a uniquely identifiable knife relevant to both conversational partners.

What has been less clear is the moment-by-moment cognitive processing by which people reconcile knowledge mismatch in language production and interpretation.  We propose a theory and computational model in which differing views of relevant knowledge, both privileged to the individual and shared with the conversational partner, are simultaneously and probabilistically considered at the earliest moments of language processing.  This approach arises from a view of communicative interaction as a process requiring continual and rapid accessing and updating of both privileged and shared knowledge.  Moreover, the probabilistic formulation emphasizes the uncertainty inherent in assessing the knowledge and beliefs of a conversational partner, and suggests that accommodating such uncertainty is therefore an inherent aspect of online language processing.

This is joint work with Daphna Heller, Associate Professor of Linguistics, University of Toronto, and Mindaugas Mozuraitis, Team Lead, Cancer Analytics at Cancer Care Ontario.


Suzanne Stevenson received a bachelor's degree in Computer Science and Linguistics from William and Mary, and master's and Ph.D. degrees in Computer Science from the University of Maryland, College Park. From 1995-2000, she was on the faculty at Rutgers University, holding joint appointments in the Department of Computer Science and in the Rutgers Center for Cognitive Science (RuCCS). She returned to the University of Toronto in July, 2000, where she is now Professor of Computer Science. She was a Visiting Professor in Linguistics at the University of California, Santa Barbara, in 2010-11 and 2015-16.

Dr. Stevenson's research is in computational linguistics (CL) and cognitive science, integrating computational theories and techniques with insights from the fields of linguistics and psycholinguistics. In cognitive science, she works on computational models of child language acquisition and adult processing, taking probabilistic approaches that learn from and adapt to the environment. Her work in CL often focuses on machine learning of semantic and syntactic information from text data, showing how linguistic knowledge or cognitive principles can help guide the learning process.

This talk is organized by Naomi Feldman