The study tested many kinds of ambiguous and unambiguous sentences with timed reading and comprehension checks. Across this wide set, cloze probability—the tendency of readers to supply a particular continuation—most reliably predicted both slower reading and more errors when ambiguity led people astray. Measures tied to verbs, surprisal, and plausibility had smaller or inconsistent effects. That pattern points toward a processing system that leans on likely continuations built from context instead of relying primarily on abstract structural preferences or raw statistical surprise.
For anyone curious about how language skills form and how comprehension can break down, these results matter. They suggest interventions to support readers could focus on strengthening contextual prediction and on teaching strategies for recognizing when an initial interpretation might be unlikely. Follow the link to read how these findings might influence education, accessibility, and tools that help people recover from misunderstanding.
Abstract
This study investigated the cognitive mechanisms underlying the processing of garden-path sentences by examining the influence of verb/structural bias, cloze probability, surprisal, and plausibility. Using self-paced reading with yes/no comprehension questions, we analyzed a structurally diverse set of 11 types of ambiguous and unambiguous sentences. Our results revealed that cloze probability was the most robust predictor of processing difficulty, significantly influencing both reaction times and response accuracy. Specifically, the likelihood of a misanalysis, as indexed by cloze scores, predicted the persistence of incorrect interpretations and reanalysis difficulty. In contrast, verb bias, surprisal, and plausibility exerted weaker or inconsistent effects, with only plausibility showing a limited interaction in the accuracy data. These findings suggest that comprehenders rely heavily on contextual cues when interpreting syntactically ambiguous input, and that reanalysis success depends not solely on structural preferences or lexical predictability but on the overall likelihood of the initial misanalysis and of the intended interpretation.
