Glicksohn, A. 1 , Magen, H. 2 & Cohen, A. 1
1 The Hebrew University of Jerusalem
2 The Hebrew University Hadassah Medical School
Statistical learning refers to the ability of human learners to capture regularities that occur over space or time. Yet sensory input often consists of many structured features (i.e., features that appear at high probability together), as well as random features. However, research to date largely explored statistical learning in a fully structured environment. Here we report a fundamental constraint on statistical learning: Randomness in the environment, even if unrelated to the learned material, dramatically impairs learning. In the structured condition participants were presented with fully structured visual input (reoccurring triplets of shapes). In the random condition the visual input was partly structured (triplets) and partly random (a set of random shapes). Lastly, in the multisensory random condition the input was multisensory, consisting of visual triplets and random syllables. Following familiarization, participants completed a familiarity test assessing knowledge of triplets. In the structured condition significant learning of triplets occurred. In contrast, in the random condition learning was completely extinguished. While modest learning occurred in the multisensory random condition, it was significantly less robust than in the structured condition. These findings suggest a constraint on learning that may have profound implications for the generalization of statistical learning to other learning domains.