From generating to violating predictions: The effects of prediction error on episodic memory

Abstract

Generating predictions about environmental regularities, relying on these predictions, and updating these predictions when there is a violation from incoming sensory evidence are considered crucial functions of our cognitive system for being adaptive in the future. The violation of a prediction can result in a prediction error (PE) which affects subsequent memory processing. In our preregistered studies, we examined the effects of different levels of PE on episodic memory. Participants were asked to generate predictions about the associations between sequentially presented cue-target pairs, which were violated later with individual items in three PE levels, namely low, medium, and high PE. Hereafter, participants were asked to provide old/new judgments on the items with confidence ratings, and to retrieve the paired cues. Our results indicated a better recognition memory for low PE than medium and high PE levels, suggesting a memory congruency effect. On the other hand, there was no evidence of memory benefit for high PE level. Together, these novel and coherent findings strongly suggest that high PE does not guarantee better memory.

How predictability and individual alpha frequency shape memory: insights from an event-related potential investigation

“…Here, we avoided specifically instructing participants to predict the upcoming items, as prior research suggests that predictive mechanisms are engaged in the presence of probabilistic regularities (Baker et al, 2014; Bar, 2007; Dale et al, 2012; Mumford, 1992). Additionally, while some studies encourage participants to actively predict (e.g., Ortiz-Tudela et al, 2023; Turan et al, 2023), if prediction is a general mechanism underlying cognition (Friston, 2010), doing so may engender a form of anticipation that differs from naturalistic predictive processing. Each image was presented for 1000ms, followed by a fixation dot in the centre of the screen that was present for 500ms.…”

“…If predictive processing regulates information uptake, the predictive model applied and the degree to which the sensory information violates this model may influence what is deemed surprising and what is subsequently encoded into memory. Importantly, whilst most studies assume that incongruent information evokes greater prediction error signals (e.g., Greve et al, 2017; Turan et al, 2023), investigating the neural underpinnings of prediction error is important when determining whether it is in fact the degree of prediction mismatch that drives the observed memory enhancement effects, or whether another factor may account for the observed patterns.…”

“…Although prediction error-eliciting information may be preferentially encoded into memory, other research demonstrates better memory for stimuli eliciting low prediction error (i.e., predictable information; Turan et al, 2023), or for both highly expected and unexpected stimuli, in contrast to moderately expected information (Quent et al, 2022). This has prompted the suggestion that the relationship between predictability and memory follows a U-shaped, nonlinear pattern, such that schema-congruent and schema-incongruent information is prioritised in memory (with schemas proposed to form the basis of predictions; Quent et al, 2022; Van Kesteren et al, 2012).…”

“…If predictive processing regulates information uptake, the predictive model applied and the predictability of the sensory information may influence what is subsequently encoded into memory. Importantly, whilst most studies assume that incongruent information evokes greater prediction error signals (e.g., Greve et al, 2017; Turan et al, 2023), an investigation into the neural correlates related to prediction may shed further light on the effect of predictability on memory.…”

“…However, to our knowledge, N400 patterns during learning have not been related to explicit memory outcomes in non-linguistic settings, and with a consideration of intrinsic neural differences. Further, few studies have investigated the accumulation of predictions and their effect on memory across longer sequences of information, with most studies focussing on pairs or triplet relationships (e.g., Long et al, 2016; Otsuka & Saiki, 2016; Rouhani et al, 2020; Sherman et al, 2022; Sherman & Turk-Browne, 2020; Turan et al, 2023). However, the temporal accuracy of ERPs may provide a means to track predictive processing in real-time and across longer associations, to understand how predictions adapt according to information accrual.…”