Neural correlates of implicit knowledge about statistical regularities

Abstract

In this study, we examined the neural correlates of implicit knowledge about statistical regularities of temporal order and item chunks using functional magnetic resonance imaging (fMRI). In a familiarization scan, participants viewed a stream of scenes consisting of structured (i.e., three scenes were always presented in the same order) and random triplets. In the subsequent test scan, participants were required to detect a target scene. Test sequences included both forward order of scenes presented during the familiarization scan and backward order of scenes (i.e., reverse order of forward scenes). Behavioral results showed a learning effect of temporal order in the forward condition and scene chunks in the backward condition. fMRI data from the familiarization scan showed the difference of activations between the structured and random blocks in the left posterior cingulate cortex including the retrosplenial cortex. More important, in the test scan, we observed brain activities in the left parietal lobe when participants detected target scenes on temporal order information. In contrast, the left precuneus activated when participants detected target scenes based on scene chunks. Our findings help clarify the brain mechanisms of implicit knowledge about acquired regularities.

Statistical Learning Signals for Complex Visual Images in Macaque Early Visual Cortex

“…Two human fMRI studies that compared regular and random sequences (Turk-Browne et al, 2009;Otsuka and Saiki, 2017), as we did here, reported the opposite effect: an increased activation for the regular sequence. The reason for this discrepancy between our monkey fMRI study and the human fMRI studies is unclear, but one important factor might be the exposure duration, which was brief (i.e., during the single fMRI session) in the human studies.…”

“…Also in rodent studies of visual statistical learning, different types of differential responses were obtained in sequence learning. One study (Gavornik and Bear, 2014) found an increased electrophysiological response to predicted gratings in mouse primary visual cortex, in line with the increased BOLD activations for regular compared to random stimuli seen in higher areas in human fMRI studies (Turk-Browne et al, 2009;Otsuka and Saiki, 2017). However, a two-photon calcium imaging study reported overall weaker visual responses for predicted compared to unpredicted stimuli in the primary visual cortex (Fiser et al, 2016).…”

“…Greater activations for random compared to regular sequences were not observed. Using a similar design, but with visual scenes as stimuli, a later study (Otsuka and Saiki, 2017) showed stronger activations to random compared to regular sequences in the left posterior cingulate cortex. Note that in these studies the temporal regularities during stimulus exposure were irrelevant to behavior.…”

Visual statistical learning based on time information.

“…It can thus be predicted that the supplementary motor area that supports time perception and the visual areas that support object recognition, such as the lateral occipital cortex, are engaged in VSL based on time information. Alternatively, considering the neural mechanisms of VSL (Karuza et al, 2017; Otsuka & Saiki, 2017, 2020; Schapiro et al, 2014; Schlichting et al, 2017; Sherman & Turk-Browne, 2020; Turk-Browne et al, 2009, 2010), I predict that brain regions, such as the medial temporal lobe, including the hippocampus, inferior parietal lobule, and inferior frontal gyrus, support VSL based on time information.…”