Roger Levy, UC San Diego

Friday, October 12th, 4pm, Stevenson Fireside Lounge

Probabilistic knowledge and locality in syntactic comprehension  

Two theoretical problems have stood at the core of psycholinguistic research in syntactic comprehension: (1) the resolution of local ambiguity; and (2) syntactic complexity, or the difficulty incurred in processing locally unambiguous structures.  This talk describes a unified treatment of these two problems through the theory of surprisal, which proposes that comprehenders rationally deploy probabilistic knowledge to yield variability in word-by-word processing difficulty that reflects a wide range of evidential information sources.  I present computational modeling and experimental studies showing how surprisal effects account for a range of both garden-path ambiguity resolution and syntactic complexity effects, and give empirical evidence for the specific quantitative relationship between subjective probability and processing difficulty -- as measured by word-by-word reading times -- proposed by surprisal theory.  For problems of syntactic complexity I compare the predictions of surprisal theory to those of theories positing a primary role of distance-sensitive locality due to memory constraints, and present new empirical data from studies on German and Russian syntactic processing that provide evidence of both surprisal and locality effects.  I close with speculation on possible ways forward toward a unified theory of probabilistic knowledge and memory constraints in incremental sentence comprehension.