Abstract
Statistical learning (SL) refers to the implicit learning of patterns among stimuli organized according to probabilistic relationships. In the visuo-temporal domain, a common paradigm involves exposure to a stream of unique shapes that, unbeknownst to the subject, has been generated from shape pairs or triplets. A variety of tasks have been employed to identify behavioral effects of SL and these are often considered to be interchangeable measures of learning. Here we operationalized SL using multiple measures of learning and found that SL may be composed of dissociable processes. In Experiment 1, subjects were exposed to a steady stream of stimuli as they performed an online task of periodically indicating whether a stimulus was present or absent, while the overall contrast was staircased to the subjects' threshold levels. Decreased response latency to the third vs. first item in triplets demonstrated learning, albeit with large variance. Experiment 2 replicated Experiment 1 and added an RSVP target detection post-task. There was a clear SL effect in target detection speed, which we used to categorize which triplets had been learned for further analysis of performance during the exposure task. Examining ‘learned’ vs. ‘unlearned’ triplets separately revealed a counter-intuitive finding: only unlearned triplets showed speeded responses in the present/absent task as a function of triplet position. These results were replicated in further experiments where we used a recognition memory post-task in place of the RSVP post-task (Experiment 3) and where we added foils to measure baseline performance (Experiment 4). Taken together, these results show that the SL process supporting speeded responses in the exposure task was different from the SL process that was expressed in the RSVP and memory post-exposure tasks. Thus, SL may consist of a family of processes that can produce dissociable effects on different aspects of behavior.
Meeting abstract presented at VSS 2013