Purchase this article with an account.
Anna Leshinskaya, Sharon Thompson-Schill; Higher Order Structure in Visual Statistical Learning. Journal of Vision 2018;18(10):262. doi: https://doi.org/10.1167/18.10.262.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
An important aspect of perception is knowing the structure of our visual world, such as what objects or events are likely to co-occur. Experiments in visual statistical learning show that participants spontaneously and implicitly learn such predictive structure (e.g., Fiser, J., & Aslin, R. N. . Unsupervised statistical learning of higher-order spatial structures from visual scenes. Psychological Science, 12, 499–504.) Here we probed whether visual statistical learning can produce higher-order knowledge: predictive relations among predictive relations. Participants performed a cover task while watching sequences composed of eight distinct events. Sequences followed certain 'rules', where a rule specified which two of the eight events were predictive. Three sequence types were shown, cued by a distinct background object. The first two sequences each followed two rules (R1 & R2 and R3 & R4), where each rule held between a unique pair of events. A third sequence contained either a consistent pairing of rules (R1 & R2) or an inconsistent pairing (R1 & R3). Critically, consistency was defined not by which events appeared, but whether the same four events participated in rules (vs. appeared randomly). Although participants had minimal awareness of these rules, their performance on forced-choice tests indicated reliable learning. Importantly, performance on the third sequence was affected by pairing consistency. Participants who saw an inconsistent pairing performed worse on the third sequence relative to their baseline (t = 2.30, p = .023), while participants who saw a consistent pairing showed no change, yielding a significant interaction (F[1, 374] = 5.33, p = .022). Thus, learners spontaneously and implicitly encoded how predictive relations themselves cohere into higher-order sets, affecting their learning of new evidence. The expectation that rules which cohered in the past will continue to cohere in the future may help us build generalizable structured models of our visual world.
Meeting abstract presented at VSS 2018
This PDF is available to Subscribers Only