Evaluating the compatibility of predictive coding as a model of linguistic predictability effects in reading

Abstract

Emerging work suggests that the computational architecture of predictive coding (PC), widely studied in visual perception, may be an appropriate model for understanding predictability effects in real-time reading (i. e., the influence of a preceding sentence context on current word processing via putative anticipatory mechanisms). We evaluated four hypotheses derived from PC models focused on: (1) bottom-up prediction error, (2) top-down sensory prediction, (3) biological plausibility, and (4) hierarchical inference. We argue that many empirical effects of predictability during reading are generally consistent with PC model hypotheses, such as ERP N400 context effects and facilitated recognition and reading times, while others are challenging to reconcile, such as the frontal positivity ERP effect to plausible prediction violations and emerging work examining brain-behavior dynamics in natural reading. Overall, we suggest that while PC applied to linguistic processing has some important explanatory utility, it likely does not reflect the sole core mechanism of the language architecture.