Abstract
The current experiments investigated how learned spatial statistics affect visual short-term memory (VSTM). In all experiments, a VSTM task was used in which a sample array was presented, followed by a 1000ms delay, followed by a test probe. Participants indicated whether the probe was present in the sample array. Sample arrays consisted of six novel shapes, similar to those used in previous studies of visual statistical learning. Structured arrays consisted of base pairs, i.e. pairs of shapes that always appeared in the same relative spatial positions (e.g. shape-A always appears above shape-B). On unstructured arrays, shapes were presented in random locations, with the restriction that the global configuration was one that could appear in the structured arrays. In experiment 1, sample arrays were presented for 2000ms. Performance was better on structured arrays, suggesting that spatial statistics can be used to enhance VSTM. Participants could also recognize the base pairs at the end of the experiment. In experiment 2, sample array inspection time was reduced to 500ms. All effects of spatial structure were eliminated, consistent with the idea that learning spatial statistics depends on inspection time. In experiment 3, participants passively viewed the structured and unstructured arrays prior to performing the VSTM task (with 500ms inspection time). As in experiment 2, performance on the VSTM task was equivalent for structured and unstructured arrays. However, unlike experiment 2, participants could recognize the base pairs at the end of the experiment. These experiments suggest that visual statistical learning does not affect the basic units of VSTM. If visual statistical learning affected VSTM's units, then performance should have been better on structured arrays in experiment 3. Thus, these findings suggest that information in long-term memory can supplement limited capacity VSTM, as long as enough time is available to access VLTM.