| |
Abstract:
Abstract: Instructed category learning tasks involve the
acquisition of a category structure through the integration of two
sources of information: explicit categorization rules provided
during initial instruction and feedback on categorization
judgements made on specific stimulus objects. Despite perfect
consistency between categorization instructions and provided
feedback, training on specific exemplars can sometimes drive
categorization behavior away from rule-following and towards a more
instance-based pattern. We summarize the results of a number of
human learning studies which have examined this interference
effect, exploring the impact of the degree of feature integration
in the stimuli, categorization rule complexity, and individual
differences in rule-following skill. We show that interference can
arise even when categorizing simple line drawings consisting of two
unintegrated features, given sufficient rule complexity. Also, the
amount of interference is shown to be inversely related to
rule-following skill. We present a connectionist model which
explains these phenomena in terms of the interaction between
activity-based processing in prefrontal cortex and weight-based
processing in sensory cortices and parahippocampal areas.
Prefrontal cortex is seen as actively maintaining a representation
of the explicitly provided rules, and this representation is seen
as modulating a posterior categorization process. Interference
arises when synaptic weight modification in the posterior system
overcomes the influence of this frontal representation. The results
of computational simulations are fit to the human learning
data.
|
