Anthony Sali¹ (), Emily Oor¹; ¹Wake Forest University

Statistical learning—the mechanism by which the visual system tracks likelihoods in the world around us—allows individuals to flexibly regulate attentional control settings, such as spatial attention shifting readiness (e.g., Sali et al., 2015), across dynamic environments. However, learned adjustments in shifting readiness do not occur in isolation and our understanding of the interaction of different statistical learning processes remains limited. Across two experiments, we investigated the processing architecture responsible for shift readiness predictions (e.g., receiving a cue to shift attention when expecting to hold) and cue stimulus identity predictions (e.g., receiving shift cue B when shift cue A is more frequent). Participants monitored one of two alphanumeric streams for an embedded cue that signaled them to hold attention at the current location or to shift attention to the opposite location and made manual responses to target digits. Experiment 1 employed four cues (two shift and two hold) such that participants could receive a high probability or low probability cue stimulus regardless of the outcome (shift versus hold) and current shift likelihood. There were substantial trial-by-trial priming effects such that exact cue stimulus repetitions were associated with shorter response times (RTs) than the other trial types. When excluding exact repetition trials, we observed additive RT costs for shift readiness and stimulus identity prediction errors. In Experiment 2, we replicated and extended these findings with twice as many shift and hold cues, allowing us to prevent any consecutive exact cue repetitions. A Bayesian analysis revealed that the likelihood of an additive relationship between the cost associated with shift readiness and stimulus identity prediction errors was approximately six times as likely as the alternative that an interaction existed. These results suggest that visual statistical learning processes governing stimulus expectations and shift readiness are distinct and constrained by a shared processing bottleneck.