The processing of mispredicted and unpredicted sensory inputs interact differently with attention

Highlights

• A joint analysis was performed on four electroencephalography (EEG) datasets.

• Mispredicted/Unpredicted processing interact differently with attention on N1.

• The mispredicted-versus-predicted contrast showed an attention-independent effect.

• The unpredicted-versus-predicted contrast showed an attention-dependent effect.

Abstract

Prediction and attention are fundamental brain functions in the service of perception. Interestingly, previous investigations found prediction effects independent of attention in some cases but attention-dependent in other cases. The discrepancy might be related to whether the prediction effect was revealed by comparing mispredicted event (where there is incorrect prediction) or unpredicted event (where there is no precise prediction) against predicted event, which are associated with different precision-weighted prediction error. Here we conducted a joint analysis on four published electroencephalography (EEG) datasets which allow for proper dissociation of mispredicted and unpredicted conditions when there was orthogonal manipulation of prediction and attention. We found that the mispredicted-versus-predicted contrast revealed an attention-independent effect of prediction suppression, whereas the unpredicted-versus-predicted contrast revealed a prediction effect that was reversed by attention on auditory N1. The results suggest that mispredicted and unpredicted processing interact with attention in distinct manners.

Introduction

Prediction and attention are two theoretical constructs thought to support perception. However, they might engage separate mechanisms. While prediction interprets sensory inputs on the basis of prior likelihood, attention prioritises stimulus processing on the basis of motivational relevance (Summerfield and Egner, 2009). This distinction is corroborated by the fact that the two mechanisms have opposite effects on early auditory processing, with prediction decreasing and attention increasing the amplitude of neuronal responses (Lange, 2013, Schröger et al., 2015).

Recently, research started to address the interaction between prediction and attention on the auditory event-related potentials (ERPs) (Bendixen et al., 2009, Winkler et al., 2012, Phillips et al., 2016). However, contradictory patterns were documented, particularly on the N1. This component is a frontocentral negative wave with a peak latency between 50 and 150 ms generated in the auditory cortex (Näätänen and Picton, 1987). Some studies reported an attention-independent effect of prediction suppression. Predicted stimuli were found to elicit smaller N1 than non-predicted stimuli independent of attention allocation (Hsu et al., 2014a, Hsu et al., 2016), suggesting that prediction and attention have additive effects. Moreover, the pattern remained when there was proper refractoriness control, excluding the possibility that the results were due to the refractory properties of neuronal populations alone (Budd et al., 1998). Meanwhile, other studies of similar design reported an effect of prediction enhancement that is attention-dependent. Predicted stimuli were found to elicit larger N1 than non-predicted stimuli under attended but not unattended condition (Hsu et al., 2014b), indicating that prediction and attention have synergistic effects.

This contradiction might be explainable if we take a closer look at the conditions that are contrasted (Feuerriegel, 2016). Generally, predicted stimuli are compared with “non-predicted” stimuli to reveal the prediction effect. In research demonstrating the attention-independent effect of prediction suppression (Hsu et al., 2014a, Hsu et al., 2016), the non-predicted stimuli were either the first tones in pairs of identical tones or the first tones in quartets of rising tones. That is, they were embedded in a context full of perceptual regularity, where participants hold rather precise prediction (i.e., high prior precision) but the sensory input does not fit with it. On the other hand, in research demonstrating the attention-dependent effect of prediction enhancement (Hsu et al., 2014b), the non-predicted stimuli consisted of random tones embedded in a stimulus stream without perceptual regularity. In this case, the formation of prediction is rather difficult (i.e., low prior precision). In short, there is a distinct difference between these two scenarios in terms of prior precision.

The distinction resonates with the idea that non-predicted stimuli should be dissociated into mispredicted and unpredicted stimuli (Arnal and Giraud, 2012, Hsu et al., 2015). Conceptually, mispredicted stimuli are associated with incorrect prediction, whereas unpredicted stimuli are associated with no precise prediction. According to the predictive coding model of perception, prediction error is the only information that needs to be communicated forward to the next higher level within hierarchical cortical network (Rao and Ballard, 1999, Friston, 2005, Friston, 2009; see Clark, 2013 for a review). Moreover, the influence of prediction error on updating future predictions is precision-weighted (Friston and Kiebel, 2009). That is, prediction error is weighted more in context of high precision (i.e., mispredicted prediction error) than in context of low precision (i.e., unpredicted prediction error). Therefore, distinct patterns of prediction effect may arise when the two types of non-predicted stimuli (associated with different precision-weighted prediction error) are compared against predicted stimuli, whose precision are high in general. The mispredicted-versus-predicted contrast would reveal prediction suppression because predicted condition is associated with the same precision weighting as mispredicted condition but has smaller prediction error. The unpredicted-versus-predicted contrast, on the other hand, would reveal prediction enhancement because predicted condition is associated with higher precision weighting than unpredicted condition. Such dissociation can also explain functional magnetic resonance imaging (fMRI) findings of enhanced activation for surprise events (when predictions are violated) and suppressed activation for neutral events (when no prediction are induced) compared to expected events (Rahnev et al., 2011, Amado et al., 2016).

It is thus possible that the inconsistency in the literature is due to the fact that the two types of non-predicted stimulus processing interact with attention in distinct manners. Here we reported a joint analysis on four published electroencephalography (EEG) datasets (Hsu et al., 2014a, Hsu et al., 2014b, Hsu et al., 2016) which allow for proper dissociation of mispredicted and unpredicted conditions when there was orthogonal manipulation of prediction and attention (see Materials and methods for details). The joint analysis is expected to elucidate whether and how the locus of attentional selection determines its gain modulation of prediction error.