Chen, Keyue; (2025) Prediction revision during language processing. Doctoral thesis (Ph.D), UCL (University College London).

Text final thesis.pdf - Accepted Version Access restricted to UCL open access staff until 1 October 2026. Download (5MB)

Abstract

Comprehenders use contextual information to generate predictions about upcoming language in real time. Recent research indicates that they can also revise these predictions when encountering cues that contradict the prediction but indicate an alternative. This dissertation investigates the mechanisms underlying prediction revision, focusing on three aspects: the cost of processing prediction-inconsistent cues, the interference of luring competitors, and the contributions of top-down and bottom-up processes. Three empirical studies were conducted in Mandarin Chinese, using experimental materials that manipulated the informativeness of prenominal cues, such as classifiers and modifiers. Anticipatory eye movements in the visual-world eye-tracking paradigm and N400 amplitude reductions in ERP experiments served as indices of prediction revision. Study 1 examined whether prediction-inconsistent cues impose processing costs that hinder subsequent revision. Results showed that early prediction violations did not disrupt subsequent semantic processing, indicating that comprehenders can rapidly and effectively revise predictions based on consecutive cues. Study 2 tested whether a luring competitor in the visual display interferes with revision. Although attention was drawn to the competitor, revision remained strongly guided by global context. Study 3 disentangled the roles of top-down global context and bottom-up lexical association in prediction revision. Local informative cues facilitated noun retrieval regardless of global plausibility, suggesting independent effects of global and local context. Overall, this dissertation demonstrates that prediction-inconsistent cues activate related lexical representations via spreading activation, thereby facilitating the processing of associated but unexpected words without incurring additional processing cost. Global context is not immediately integrated with local cues to constrain revision to globally plausible alternatives; rather, it provides independent support for plausible candidates. Consequently, locally compatible but globally implausible competitors may modestly interfere with ongoing processing. These findings indicate that prediction revision is a dynamic process shaped by both immediate cues and broader context, with top-down and bottom-up processes operating in parallel.

Download activity - last month

Export as JSONCSVXML

Download activity - last 12 months

Export as JSONXMLCSV

Downloads by country - last 12 months

Export as JSONXMLCSV

Archive Staff Only

View Item