Abstract
The visual system is an efficient statistician, extracting statistical summaries over sets of objects (statistical summary perception; SSP), as well as statistical regularities between individual objects (statistical learning; SL). While both of kinds of statistical processing have been studied extensively, their relationship is currently unknown. Here we examine whether, and how, SSP and SL influence each other. Observers were familiarized to displays that were constructed, without their knowledge, from spatial pairs of oriented lines that could only be learned based on statistics of co-occurrence. In Experiment 1, we examined whether SSP influences SL for these pairs by manipulating the task performed during familiarization. Fifty-four observers were randomly assigned to one of three tasks: passive viewing, summary (judging mean line orientation vs. vertical meridian), or control (detecting the presence of duplicate orientations). SL, expressed in a subsequent forced-choice familiarity test (chance = 50%), was reliable in passive viewing and control conditions (56.3% and 65.3%, respectively), but significantly weaker in the summary condition (48.6%). These results do not reflect dual-task interference per se (because of the control condition), but suggest that SSP in particular hinders SL. In Experiment 2, we examined whether SL influences SSP by manipulating the presence of statistical regularities during the summary task. Twenty new observers were randomly assigned to one of two conditions: structured (summary condition from Experiment 1), or random (grids generated like before, but positions scrambled on each trial). Performance on the summary task (chance = 50%) was reliable in structured and random conditions (56.6% and 64.0%, respectively), but significantly worse for structured. These results suggest that the presence of statistical regularities (even if not ultimately learned) hinders SSP. In sum, we find bidirectional interference between extracting statistical summaries and learning statistical relationships. These findings are consistent with competition for shared statistical resources and/or dependence on different modes of attention.