Statistical Learning without Attention
43.4089, Monday, 18-May, 8:30 am - 12:30 pm, Pavilion
Feitong Yang1, Jonathan Flombaum1; 1Department of Psychological and Brain Sciences, Johns Hopkins University
We sought to investigate the role of attention in statistical learning, an area where current results conflict. Given a stream of shapes including two different colors, and instructed to attend one of the colors (via a cover task), will observers learn statistical regularities associated with the unattended color? Following previous studies, we employed a reaction time (RT) test following encoding. Speeded responses are made to a target shape (on each trial) embedded within an RSVP stream, with learning demonstrated as an RT benefit for second and third triplet items. However, typical procedures repeat the same triplets as fillers and test items, making learning during testing possible. We therefore conducted an experiment with only a test phase (i.e. no incidental exposure), and we found significant RT benefits consistent with statistical learning, even in the first 48 of 96 test trials. These results demonstrate that statistical learning can take place rapidly during the course of procedures that are at times employed to diagnose prior learning. We thus returned to the question of attention with a modified test procedure. In addition to a set of eight triplets shown during incidental exposure, we generated a place-holder-set of 12 additional shapes. Each test trial then included a target shape from one of the learning triplets. It appeared embedded appropriately within its triplet, but with that triplet embedded within a larger set including nine of the 12 place-holder-items. After confirming a lack of statistical learning during the test phase (i.e. without pre-exposure), we used it as the test component for the attended and unattended color experiment described initially. We found significant learning effects for attended and unattended shapes. In addition to furnishing an updated RT test, these results demonstrate the robustness of statistical learning, which arose rapidly and for unattended stimuli.