Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Hierarchical Predictive Processing during Naturalistic Reading
Poster Session C, Saturday, September 13, 11:00 am - 12:30 pm, Field House
Lucas Yiu Hei Chan1, Olaf Dimigen2, Urs Maurer1,3; 1Department of Psychology, Chinese University of Hong Kong, Hong Kong, 2Faculty of Behavioural and Social Sciences, University of Groningen, Groningen, The Netherlands, 3Brain and Mind Institute, Chinese University of Hong Kong, Hong Kong
Predictive processing has been a major framework for studying cognitive mechanisms. In the literature of language comprehension, studies have demonstrated that the brain elicits prediction error signals specific to various lexical and sub-lexical word properties. Recent evidence also showed that these predictions are likely organized hierarchically during naturalistic listening. Naturalistic reading, on the other hand, has received less attention in the field. We therefore tested the hierarchical predictive processing proposition in the domain of naturalistic reading comprehension. We expected to see three types of evidence as a result of a hierarchical predictive mechanism - 1) distinct prediction error signals at each level of the hierarchy, 2) a temporal order of prediction error signals from lower to higher levels of the hierarchy, and 3) higher-level predictions constraining lower-level predictions. To test our hypotheses, we asked participants to read fictitious story while their EEG and eyetracks were recorded. Using time-resolved regression, we modelled their regression fixation related potentials (rFRPs) with word surprisal, frequency, and orthographic prediction error (oPE) as regressors, each representing the prediction error signal on a specific level of the hierarchy. Cluster-based permutation showed that each of these regressors elicited a significant cluster in the rFRP. Word surprisal captured an N400 component; word frequency explained variance in the N1 and N250 components; and oPE explained variance in the narrow time range of N1 onset. This result suggested that each of these prediction error signals has a distinct topographic and temporal pattern which constituted the first type of evidence we expected. Secondly, we sought to investigate the temporal order between these prediction error signals. For each effect, we identified peak effect latencies of each individual participants by finding the maximum correlation between the average peak effect topography and participant's topography at each time point of the rFRP. We found that the effect oPE was elicited significantly earlier than the effects of surprisal and frequency. No significant differences were found between surprisal and frequency effects. Hence, this provided evidence for a temporal order of prediction error signals where the sub-lexical signals (i.e., oPE) was elicited earlier than the lexical signals (i.e., surprisal and frequency). Lastly, we compared regression models differed on how oPE was calculated to test whether higher-level lexical predictions constrained lower-level orthographic predictions. We derived context-based and frequency-based oPE. The former weighted orthographic predictions with contextual word probability and the latter weighted them with word frequency. Both oPE represented hierarchical processes in the sense that orthographic predictions were dependent on higher-level word probabilities. Cross-validated R2 showed that the model with context-based oPE performed significantly better than the one with frequency-based oPE and the non-hierarchical model. This suggested that the brain uses prior context to formulate orthographic predictions. All in all, our study demonstrated that naturalistic reading engages a hierarchical predictive process as proposed by the predictive processing framework.
Topic Areas: Reading,