Chunking and implicit pattern learning in a serial reaction time task
Abstract
Several reports of implicit pattern learning have appeared in recent years and are briefly reviewed. In the experiments reported here, the experimental task was a serial reaction time task in which subjects were required, on each of several successive trials, to indicate by pressing a key in which of four locations on a computer screen a target character appeared. The basic finding is that subjects' performance on this task is facilitated by the imposition of patterns on the serial order of the stimuli in the task, even though they often are unable to describe the patterns. For example, performance is better when a repeating pattern of ten trials determines the sequence of target locations than when the sequence is random. The present experiments investigate the hypothesis that implicit pattern learning relies on chunking (organization) of the sequence of trials specifically experienced by the subjects. Improvement of performance with practice is assumed to be specific to the chunks experienced during training; performance is thought to be better on a patterned sequence than on a random sequence because in the patterned sequence, there are fewer chunks to practice and chunks are practiced more frequently and are repeated with fewer intervening trials. Four experiments investigated the chunking hypothesis by (1) manipulating sequence structure so that the number of chunks encountered by the subjects and the frequency and spacing of the repetitions of those chunks were varied and (2) manipulating the temporal grouping of the sequence so that grouping was consistent or varied over repetitions of the pattern. The results supported the chunking hypothesis. As the number of chunks in a pattern decreased, greater learning was observed. Learning was also observed when the sequence of trials alternated between trials ordered by a pattern and trials ordered randomly. When the sequence was ordered according to probabilistic constraints of varying orders of approximation to a pattern, learning occurred in relatively low orders of approximation, and increased as the approximation increased. Also, learning was greater for consistent than varied grouping. The implications of the results are discussed, along with possible directions for future research.
Degree
Ph.D.
Advisors
Roediger, Purdue University.
Subject Area
Psychology|Experiments
Off-Campus Purdue Users:
To access this dissertation, please log in to our
proxy server.