hide
Free keywords:
-
Abstract:
Interaction with our continuously changing environment relies on anticipating timing of events, enhancing information processing efficiency. Abundant research has investigated temporal prediction in deterministic environments such as isochronous rhythms, where the presumed mechanism is oscillatory entrainment to external rhythms. However, in everyday life, continuous streams lack fully-deterministic temporal regularities. Previous research of temporal prediction in uncertain environments has focused on isolated intervals, suggesting a Distributional-Learning (DL) model. Alternatively, Recent studies suggest that oscillatory entrainment could explain predictions in non-isochronous streams. However, in non-deterministic streams, if and under which conditions either of these mechanisms drives prediction is unclear. To address this, we combined computational modeling of these two mechanisms, oscillatory entrainment (OE) and a Sequential Bayesian-Updating (a DL model), and human behavioral experiments. First, we examined the models’ dynamic predictions in streams with zero, low or high inter-stimuli interval (ISI) variability. We found that both models lead to overlapping predictions in zero ISI variability. Moreover, while prediction certainty of OE model is affected nonlinearly by degree of variability and the specific order of stream intervals, DL model is linearly dependent on ISI variability. Next, we used the models generatively to create streams with differential temporal predictions by these two mechanisms, and presented targets at either timepoint to participants conducting a speeded response task. Participants’ behavior followed predictions of the DL model, though only in low degrees of variability. Overall, these results highlight the inherent differences between OE and DL mechanisms, and the superiority of DL-based temporal predictions in uncertain continuous environments.
