Formal and Neural Models of Pragmatics
Instructor: Jackie Chi Kit Cheung
Time: Tuesdays and Thursdays, 11:30am – 1:00pm
Location: Burnside 719A
Office hours: TBD
Office hours location: McConnell 108N
Large language models have displayed an impressive array of skills and behaviours. They are now in deployment in many applications where they directly interface with human users. This motivates the need for further development of models and evaluations related to their ability to process implicit or intended meaning in context, which falls under the field of pragmatics in linguistics.
In this course, we will examine computational models of pragmatics and how NLP systems have been empirically evaluated for their pragmatic reasoning ability. We will discuss classical theories of formal semantics and pragmatics, as well as more recent statistical and neural models. Topics to be covered may include:
- Dynamic semantics (e.g., File Change Semantics; Discourse Representation Theory)
- Reference and coreference; anaphora
- Implicature and presupposition
- Common sense knowledge representation and reasoning
- Discourse structure; coherence modelling
- Information structure; definiteness
Prerequisites: One course on NLP (e.g. COMP 550) or permission of the instructor
The course format will consist of lectures delivered by the instructor, mixed with paper presentations and discussions led by the students. The student's grades will be based on i) paper presentations and discussions, and ii) a final research project.
- Classes start January 4th, 2024. Course discussions will take place on the Ed platform. This class is anticipated to be an in-person only class.