Modeling Psycholinguistic Processes as Sequential Decision Processes
Learnable from Experience

The talk will discuss ways in which (deep) reinforcement learning
methods that induce sequential decision processes from trial-and-error
experience could be used to shed light on the modeling and
learnability of some psycholinguistic processes. We’ll use left-corner
chart parsing as our main running example (joint work with Rohan
Pandey and Maximilian Alfano-­Smith), but we will outline other
applications of these methods, e.g., rule utility learning in
production-based cognitive models of psycholinguistic processes (joint
work with Jakub Dotlacil), and task effects (self-paced reading vs.
Maze) in the real-time processing of polysemes vs. homonyms (joint
work with Jack Duff and Amanda Rysling).