Uncertainty regulates statistical learning mechanisms involved in cue processing : behavioral, eye-tracking, and neurophysiological Evidence

Abstract

Statistical learning, i.e., an incidental acquisition of environmental patterns based on input statistical properties (e.g., co-occurrence probabilities), is driven by implicit (subconscious) and explicit (conscious) systems. Nevertheless, it remains unknown which factor determines the engagement of these statistical learning mechanisms and related cognitive processes. Employing behavioral (Chapter 2), eye-tracking (Chapter 3), and electroencephalography (EEG) techniques, the present thesis examined whether and, if so, how input statistical properties (i.e., uncertainty) regulate the cognitive processing of associative patterns during visual statistical learning. Specifically, a novel probabilistic cueing-validation paradigm was created, where one set of stimuli (i.e., targets) appeared after another set of stimuli (i.e., predictive cues) with high- (75%) or low- (25%) transitional probability (TP), which induced certainty or uncertainty, respectively. After these cue-target sequences, high-, low-, and zero-probability associated cue objects recurred, serving as probes of associative patterns. Chapter 2 examined whether and, if so, how preceding target TP impacts internal representations of post-target cue objects. Results demonstrated that high-probability associated cues exhibited suppressed representations after high-TP targets, regardless of learners’ eventual awareness of cue-target associations. In contrast, after low-TP targets, participants who recognized cue-target associations above chance-level, showed enhanced representations of these cues as learning progressed. These findings suggest that learners may adapt to high-probability associations when perceiving certainty, but exploit high-probability associations to cope with uncertainty. To further investigate input-regulated and individual-variant cognitive processes, Chapter 3 examined whether and, if so, how participants’ statistical learning performances, assessed by implicit (i.e., target detection) and explicit (i.e., target recognition) tasks, influence their attentional biases on cue objects after high- and low-TP targets. Linear mixed-effects modeling analyses revealed that good (1.5 SD above the mean) implicit learners showed faster first fixation latency and longer total dwell time on low- compared to high-probability associated cues; whereas after low-TP targets, good explicit learners exhibited faster first fixation latency on high-probability associated cues as learning progressed. These results indicate dissociated attentional strategies, i.e., exploring low-probability associations and exploiting high-probability associations, driven by implicit and explicit mechanisms, respectively. Next, to shed light on the neural mechanisms underlying the processing of associative patterns, Chapter 4 examined oscillatory correlates of probabilistic associations after high- and low-TP targets. Time-frequency analyses revealed a desynchronization in the beta (10-39 Hz) frequency band in the parietal region, attenuated by high-probability associated cues after high-TP targets. In addition, synchronization in the theta (4-10 Hz) frequency band was observed in the frontal region with enhanced power on these cues after low-TP targets. Additionally, low-probability associated cues elicited a lower parietal theta synchronization after low- compared to high-TP targets. Moreover, individual differences in oscillatory effects predicted individual differences in the magnitude of the preceding target-elicited statistical mismatch negativity component (sMMN, low- vs. high-TP). These results imply that certainty-driven memory consolidation, uncertainty-driven memory retrieval, and attentional disengagement from error signals may occur during pattern acquisition. In sum, during visual statistical learning, input statistical properties regulate the engagement of multiple cognitive mechanisms, manifesting as a certainty-driven, implicit, adapted process and a set of uncertainty-driven, explicit, error-coping processes.