Attention and probabilistic sequence learning

Abstract

Limitations of using fixed sequences of events in studies of learning in the sequential reaction-time task led us to develop a probabilistic version of the task. When sequences occur probabilistically, transitions usually follow a sequence, but with some small probability, events occur out of sequence. This variation on the paradigm provides new evidence associated with manipulations of attentional load. Most notably, single-task learning leads to particularly high error rates on improbable transitions, suggesting anticipation of the sequence. Dual-task learning shows sensitivity to the sequence (by reaction-time differences to probable and improbable transitions), but without inflated errors on improbable transitions. Sensitivity to the sequence and anticipatory errors disappeared when participants transferred from single-task learning to dual-task conditions, suggesting that what is learned with single-task practice cannot be applied under conditions of limited attention. When learners transferred from dual- to single-task conditions, sensitivity of RT to the sequence increased but anticipation errors remained the same, suggesting that attentional load limits performance, but not learning. Qualitative differences in performance result from variations in attentional resources, which may reflect different learning processes.