Predictability in Language Comprehension: Prospects and Problems for Surprisal

Abstract

Surprisal theory proposes that a word's predictability influences processing difficulty because each word requires the comprehender to update a probability distribution over possible sentences. This article first considers the theory's detailed predictions regarding the effects of predictability on reading time and N400 amplitude. Two rather unintuitive predictions appear to be correct based on the current evidence: There is no specific cost when an unpredictable word is encountered in a context where another word is predictable, and the function relating predictability to processing difficulty is logarithmic, not linear. Next, the article addresses the viability of the claim, also associated with Surprisal, that conditional probability is the “causal bottleneck” mediating all effects on incremental processing difficulty. This claim fares less well as conditional probability does not account for the difficulty associated with encountering a low-frequency word or the difficulty associated with garden path disambiguation. Surprisal provides a compelling account of predictability effects but does not provide a complete account of incremental processing difficulty.