May 23

Society for Neuroscience 2012 (New Orleans, LA)

Statistical learning in perceptual-motor sequences and planning effects in performance (PDF)

Daniel J. Sanchez1, Hristo Bojinov2, Patrick D. Lincoln3, Dan Boneh2, Paul J. Reber1
1: Northwestern University (Psychology), 2: Stanford University (Computer Science), 3: SRI (Computer Science Laboratory)

The performance improvements that underlie motor skill expertise rely on implicit learning within cortico-striatal circuits connecting the basal ganglia to motor cortex and motor planning areas. Repetition-based learning is hypothesized to develop in a bottom-up manner, such that learning of a complex sequence of motor movements is built up from smaller sub-sequences that are learned first. Using the Serial Interception Sequence Learning (SISL) task, this hypothesis was tested by training participants with long sequences and then testing whether their sequence knowledge was expressed on a test containing short sub-sequences from the trained sequence embedded among untrained segments. In the SISL task, cues scrolled vertically down a computer monitor towards six target zones, corresponding horizontally to locations on a keyboard. Multiple cues were simultaneously visible on the screen, allowing for rapid continuous performance and planning of subsequent responses. Participants attempted to intercept the cues by pressing the corresponding keyboard button as the cue passed through the target zone. Using an online SISL applet, 66 participants received 3780 trials of training on a covertly-embedded 30-item repeating sequence, followed by tests of sequence fragments of 3, 4, and 5 items in length. The 30-item sequences were structured so that all 1- and 2-item sequences appeared equally often, thereby making 3 items the smallest statistically-predictable fragment size. Participants exhibited sequence knowledge for all tested lengths, even the shortest 3-item fragments, with the most robust performance enhancement found for the center items in the 4-and 5-item fragment lengths (3rd item, and 3rd and 4th items, respectively). Enhanced performance on the first 2 items in a fragment was not anticipated because these items are not predictable from preceding cues. Unexpectedly, the last item in a fragment was also found to be performed at a low rate in spite of it being preceded by well-performed items in the sequence. This effect appears to reflect a multi-response planning process and interference from the next upcoming item when it does not follow the trained repeating sequence. A consequence of this effect is that sequence knowledge is only weakly expressed for 3-item fragments, even though the 3rd item in a 4- or 5-item fragment is performed at high levels of success. We hypothesize a bottom-up statistical learning mechanism to be an important part of perceptual-motor sequence learning; however, expression of this sequential knowledge can be affected by task constraints such as the need to plan a response to upcoming items varying from the practiced sequence.