| |
Abstract:
Abstract: Performance in an artificial grammar learning task
does not depend on explicit memory for exemplars used during
training. Several aspects of the grammar may be learned implicitly,
including the frequencies of different letter bigrams and trigrams
(chunk strength) and rules governing legal letter strings. These
different forms of implicit learning are likely to have different
neural substrates. We sought to examine brain activity during an
artificial grammar classification task in which grammatical and
nongrammatical items were balanced with respect to chunk strength.
Four subjects participated in an fMRI study in which they viewed 46
letter strings formed according to the artificial grammar rules.
After being positioned in the scanner, subjects were presented with
64 letter strings, half of which were grammatical and half were
nongrammatical. The strings were presented for 1 second each, and
the subject classified them as grammatical or nongrammatical. There
was a 14 second delay between stimuli to allow hemodynamic
recovery. In all subjects there was a significant increase in
activity in the left putamen for grammatical vs. nongrammatical
items. In 3 of 4 subjects, there was a significant increase in
activity in left caudate and right anterior cingulate. Activity in
dorsolateral prefrontal cortex did not differ between the
grammatical and nongrammatical items. These results suggest that
the basal ganglia may be involved in the implicit learning of rules
based on covariations across trials.
|
