A Predictive Coding Model of the N400

Abstract

The N400 event-related component has been widely used to investigate the neural mechanisms underlying real-time language comprehension. However, despite decades of research, there is still no unifying theory that can explain both its temporal dynamics and functional properties. In this work, we show that predictive coding – a biologically plausible algorithm for approximating Bayesian inference – offers a promising framework for characterizing the N400. Using an implemented predictive coding computational model, we demonstrate how the N400 can be formalized as the lexico-semantic prediction error produced as the brain infers meaning from linguistic form of incoming words. We show that the magnitude of lexico-semantic prediction error mirrors the functional sensitivity of the N400 to various lexical variables, priming, contextual effects, as well as their higher-order interactions. We further show that the dynamics of the predictive coding algorithm provide a natural explanation for the temporal dynamics of the N400, and a biologically plausible link to neural activity. Together, these findings directly situate the N400 within the broader context of predictive coding research, and suggest that the brain may use the same computational mechanism for inference across linguistic and non-linguistic domains.

Predictive Coding explains the dynamics of neural activity within the left ventromedial temporal lobe during reading comprehension

“…Similar to predictive coding, IAC architectures allow for an unrestricted top-down ow of information across multiple, hierarchically-organized levels of representation, which could pre-activate lower-level lexico-semantic representations that are subsequently con rmed by expected inputs. If one further assumes that these expected inputs induce a smaller shift in state at the lexico-semantic level than unexpected inputs, this type of architecture could also explain why, in the present study, the ventromedial temporal lobe both reinstated prior lexico-semantic predictions to expected input, and generated a larger univariate response to unexpected inputs (although we note that, unlike predictive coding, this does not explain how the brain computes this change in state, or why such a change in state would result in a smaller evoked response 24,25 ).…”

“…This is the classic N400 effect 19,22 , which, in plausible sentences, localizes to leftlateralized temporal regions that support lexico-semantic processing, including the left ventromedial temporal lobe 7,23 . Within predictive coding, this larger N400 to unexpected words is interpreted as an enhanced lexico-semantic prediction error [9][10][11][12]24 . This error is computed by lexical and semantic "error units" when prior top-down lexico-semantic predictions fail to suppress new unexpected lexico-semantic information being inferred within "state units" 24 .…”

“…Within predictive coding, this larger N400 to unexpected words is interpreted as an enhanced lexico-semantic prediction error [9][10][11][12]24 . This error is computed by lexical and semantic "error units" when prior top-down lexico-semantic predictions fail to suppress new unexpected lexico-semantic information being inferred within "state units" 24 . This error plays a crucial role in an optimization algorithm used to infer the unexpected word's meaning from its orthographic form.…”

Can Prediction Error Explain Predictability Effects on the N1 during Picture-Word Verification?

“…We note that exploratory analyses at the typical latency of the N400 revealed a pattern which also appears to run counter to a simple predictive coding account of predictability effects. This is seemingly inconsistent with interpretations of the N400 as indexing prediction error (Bornkessel-Schlesewsky & Schlesewsky, 2019; Eddine et al, 2023; Rabovsky & McRae, 2014). At the lowest level of predictability, we observed greater N400 amplitudes for picture-congruent words, than for picture-incongruent words.…”

“…For example, consider the well-researched N400 ERP component, generally recognised since its initial identification as capturing activity related to semantic processes (Kutas & Federmeier, 2011; Kutas & Hillyard, 1980). The N400 shows sensitivity to word- and sentence-level surprise or predictability (Delaney-Busch et al, 2019; Lau et al, 2013; Lindborg et al, 2023; Mantegna et al, 2019; Van Petten & Kutas, 1990), in a manner that may be consistent with predictive coding (Bornkessel-Schlesewsky & Schlesewsky, 2019; Eddine et al, 2023; Rabovsky & McRae, 2014). Similar interpretations have been made of other signals, as capturing prediction errors for phonological, semantic, or syntactic representations (Fitz & Chang, 2019; Gagnepain et al, 2012; Van Petten & Luka, 2012; Ylinen et al, 2016, 2017).…”