Schedule of Events | Symposia

Evaluating Hierarchical and Production-Based Prediction Models: Evidence from a Prediction-Production EEG Study

Poster Session E - Monday, March 31, 2025, 2:30 – 4:30 pm EDT, Back Bay Ballroom/Republic Ballroom

Agnes (Yang) Gao¹ (aygao@ucdavis.edu), Matthew Traxler, Tamara Swaab; ¹University of California Davis

This study tested assumptions from two theories of prediction during language processing: the hierarchical prediction model and production-based prediction models. The hierarchical model posits that comprehension involves top-down predictions from higher-level semantic representations to lower-level form-based processing, with confirmed predictions reducing processing effort and failed predictions triggering error signals for learning. In contrast, production-based models argue that comprehension relies on covert production simulations, with form-based features pre-activated through the production system and stronger engagement when tasks require verbal production. To examine these predictions, participants (N = 30) completed 480 prime-target trials while their EEG was recorded and engaged in a prediction-production task. After reading a prime word, participants verbally produced a word before seeing the target word they predicted based on the prime’s meaning. Trials were categorized as same-related (prediction identical to target), different-related (prediction semantically related but different), and different-unrelated (prediction semantically unrelated to target). ERN was found only for different-related trials, consistent with the hierarchical model’s prediction that error signals occur only for semantically plausible mismatches. P2 facilitation in same-related trials suggests pre-activation of word-form features, supporting both hierarchical and production models. Importantly, N400 effects showed reduced amplitudes for same-related trials but pervasive concreteness effects across all conditions, aligning more closely with production models by indicating deep semantic processing induced by the production task. Overall, the findings suggest that while error-based learning aligns with hierarchical predictions, and that the production task’s influence on semantic processing makes production-based models more compatible with the observed results.

Topic Area: LANGUAGE: Lexicon