| When exposure to a task facilitates performance without producing corresponding changes in reportable knowledge, learning is said to be implicit. Event-related brain potentials (ERPs) were recorded as subjects practiced an implicit, structured sequence learning task. An apparent negativity--peaking at 400 ms post-stimulus, largest over centro-parietal scalp locations--emerged with practice to violations of sequential expectancy. This effect resembles both a "delayed" N200 and an N400 component. To account for the latter possibility, a statistically-based theory of the N400 is proposed, then tested in a pair of connectionist simulations. However, later analyses revealed that this effect was due mostly to a delay in the response positivity as a function of expectancy, rather than to the addition of negative energy in the 400 ms latency range.;The ERP correlate of performance was subsequently used as a dependent measure in order to address the two enduring controversies regarding implicit learning: what is being learned under these conditions, and does such learning really occur outside of awareness? In one experiment, subjects were trained on a subset of the sequential stimuli, then tested on the complete sequence. No significant behavioral or electrophysiological effects of transfer were observed, suggesting that subjects had inferred local knowledge of the distributional properties of the learning domain, rather than inducing the general rules underlying the stimulus regularities.;In other experiments, the paradigm was modified so that subjects had explicit information available to them about the structure of the stimulus sequences. The ERPs elicited by violations of implicit and explicit expectancy appeared to reflect the modulation of a single brain process, with the magnitude of a late positivity reflecting varying amounts of concomitant awareness. This was true regardless of whether explicit expectancies were established via a data-driven technique, adding explicit cues to the display, or via a conceptually driven technique, simplifying the grammar to the point that it could be taught to subjects prior to the experiment. However, the application of the adjacent response (ADJAR) filter to correct for overlap in stimulus- and response-locked componentry revealed qualitatively different effects as a function of awareness, suggesting separable underlying mechanisms. |
