BR A I N R ES E A RC H 1 1 7 9 ( 2 00 7 ) 8 9 –1 05

a v a i l a b l e a t w w w. s c i e n c e d i r e c t . c o m

w w w. e l s e v i e r. c o m / l o c a t e / b r a i n r e s

Research Report

Neural mechanisms for learning actions in context

Phan Luu a,b,⁎, Don M. Tucker a,b , Roy Stripling c

a
Electrical Geodesics, Inc., 1600 Millrace Dr. Suite 307, Eugene, OR 97403, USA
b
Department of Psychology, University of Oregon, Eugene, OR 97403, USA
c
US Naval Research Laboratory, 4555 Overlook Ave. SW, Washington, DC 20375, USA

A R T I C LE I N FO AB S T R A C T

Article history: The transition from actions that require effortful attention to those that are exercised
Accepted 23 March 2007 automatically reflects the progression of learning. Full automaticity marks the performance
Available online 5 June 2007 of the expert. Research on changes in brain activity from novice to skilled performance has
been consistent with this behavioral characterization, showing that a highly practiced skill
Keywords: often requires less brain activation than before practice. Moreover, the decrease in brain
Event-related potential activity with practice is most pronounced in the general or executive control processes
ERP mediated by frontal lobe networks. Consistent with these human cognitive neuroscience
Learning findings, animal neurophysiological evidence suggests that two elementary learning
Medial frontal negativity systems support different stages of skill acquisition. One system supports rapid and
MFN focused acquisition of new skills in relation to threats and violations of expectancies. The
P300 other involves a gradual process of updating a configural model of the environmental
Expertise context. We collected dense array electroencephalography as participants performed an
Action arbitrary associative (“code learning”) task. We predicted that frontal lobe activity would
decrease, whereas posterior cortical activity would increase, as the person gains the
knowledge required for appropriate action. Both predictions were confirmed. In addition, we
found that learning resulted in an unexpected increase in activity in the medial frontal lobe
(the medial frontal negativity or MFN). Although preliminary, these findings suggest that the
specific mechanisms of learning in animal neurophysiology studies may prove informative
for understanding the neural basis of human learning and executive cognitive control.
© 2007 Published by Elsevier B.V.

1. Introduction
When experts become skilled in a given task domain, they
perform routine tasks automatically, thereby freeing up
cognitive resources for more executive tasks, such as organiz-
ing situational awareness (Fitts, 1964). Psychological studies of
expert cognition have emphasized the importance of controlled
processes, which are limited by the capacity of working me-
mory, require active attention, and can be directed con-
sciously in new task situations (Schneider and Shiffrin, 1977;
Shiffrin and Schneider, 1977). These are contrasted with au-
tomatic processes, which result from learning and are not
strongly limited by the capacity of working memory, and
which, when engaged in the appropriate context, may be
carried out unconsciously with minimal distraction of con-
scious attention (Chein and Schneider, 2005).
NeuroImaging studies have suggested there may be direct
neural correlates of the reduced demands for controlled pro-
cesses, as evidenced by decreased demands on brain activity
resulting from increasing practice with task performance.
Given that frontal lobe activity is thought to be particularly
important to goal representations and providing control-

⁎ Corresponding author. Electrical Geodesics, Inc., 1600 Millrace Dr. Suite 307, Eugene, OR 97403, USA. Fax: +1 541 687 7963.
E-mail address: pluu@egi.com (P. Luu).

0006-8993/$ – see front matter © 2007 Published by Elsevier B.V.

doi:10.1016/j.brainres.2007.03.092
90 BR A IN RE S E A RCH 1 1 79 ( 20 0 7 ) 8 9 –1 05
related outputs (Miller and Cohen, 2001), it is theoretically
important that both meta-analysis of fMRI studies and new
experiments have suggested a specific decrease in frontal lobe
activity (bilateral dorsal frontal, left ventral prefrontal, anterior
cingulate cortex, left insular regions) as participants become
more practiced in task performance (Chein and Schneider, 2005).
In attempting to link the more elementary neurophysiology
of action regulation to the development of expert control of
human cognition, we looked to the literature on neurophysio-
logical models of animal learning. In research on discriminant
learning in mammals, two cortico-limbic-thalamic circuits
have been identified, each providing a unique strategic control
on the learning process (Gabriel et al., 2002). The first system is
responsible for rapid discriminant learning (both avoidance
and conditional). This system includes the anterior cingulate
cortex (ACC), amygdala, and mediodorsal nucleus of the
thalamus. The unique properties of the fast learning system,
e.g., its contribution to overcoming habitual responses, led
Gabriel and colleagues (2002) to suggest this circuit is integral
to what has been called the executive control of cognition
(Posner and DiGirolamo, 1998). Bussey and colleagues (2001)
have shown that the ventral and orbital prefrontal cortices
should also be included as part of this fast learning system.
A second circuit, centered on the posterior cingulate cortex
(PCC) and anterior thalamic nucleus, is involved in the late
stages of learning (Bussey et al., 1996; Keng and Gabriel, 1998).
In these late stages, a contextual model is formed, and minor
changes that are consistent with the contextual model can be
made with minimal attentional demands. This late stage
learning process can be described as forming the basis of
habits (Bussey et al., 1996).
If this neurophysiological model is correct, learning is
achieved by circuits with qualitative strategic biases, toward
either (1) gradual updating of a valued context model or (2)
rapid, focused changes of associations under conditions of
threat or context violation (Tucker and Luu, 2006). In extend-
ing this model to expertise in human learning, the rapid,
focused learning system would be required as participants
learn new arbitrary associations, such as in visuomotor
mapping of an arbitrary visual stimulus to a particular
response (Bussey et al., 2001; Wise and Murray, 2000).
Researchers studying arbitrary visuomotor mapping (also
referred to as conditional visuomotor mapping) regard the
underlying processes as indicative of the remarkable cap-
ability of mammals to associate any arbitrary stimulus with a
motor response, and thus to behave appropriately in any given
context.
Functional neuroimaging studies of hemodynamic res-
ponses during arbitrary visuomotor association learning have
identified the involvement of the anterior cingulate sulcus,
parahippocampal gyrus, caudate nucleus (i.e., dorsal stria-
tum), inferior frontal gyrus, middle temporal gyrus, dorsal
premotor cortex, and parietal cortex (Toni and Passingham,
1999; Toni et al., 2001; Wise and Murray, 2000). These regions
appear to overlap substantially with the neural networks
implicated in controlled cognitive processes by Chein and
Schneider (2005), and certain of these structures are closely
related to the fast learning cortico-thalamic-limbic circuit
implicated by Gabriel and colleagues in the early stages of
discriminant learning.
Although we have emphasized the importance of the ACC
in action monitoring (Luu and Tucker, 2003), many of the
human findings of error-related negativities (response-locked
ERNs) or medial frontal negativities (stimulus-locked MFN or
N2) have shown strong responses in medial frontal cortex
when the subject's expectancies are violated. These results
may be consistent with more elementary learning processes;
discrepancy or conflict detection is a requirement for beha-
vioral adjustments. Indeed, it is through the detection of
discrepancies between what is expected, given a particular
action, and what actually happened that new learning occurs
(Rescorla and Wagner, 1972). According to an influential
theory of medial frontal control of cognitive conflict, the ERN
reflects the activity of a learning system that relies on
prediction errors (Holroyd and Coles, 2002).
In the present study, we presented participants with an
arbitrary associative (“code learning”) task, in which they
needed to discover an arbitrary mapping of stimuli (digits or
spatial location) with key presses of the correct hand and
finger, or with no response. We hypothesized, broadly, that
the learning process would engage frontal lobe activity
implicated in controlled cognitive processes. More specifically,
in line with the neurophysiologic model of action regulation,
we hypothesized that the demands on rapid learning early in
the task would engage activity of the anterior ventral network.
In contrast, we hypothesized that as participants learned this
task, the consolidation of a cognitive context for performance
would be indexed by an increasingly robust P300 or Late
Positive Complex (LPC), generated in medial temporal, poster-
ior cingulate, and parietal cortices, thus reflecting the more
automatic process typical of expert performance. For both the
frontal, ventral contribution to early learning, and the poster-
ior, dorsal contribution to late learning, we hypothesized that
the digit code task would result in greater activity in left
hemisphere networks and the spatial code task would result
in greater activity in right hemisphere networks.
2. Results
2.1. Behavioral data
The following factors were considered in repeated measures
ANOVA models: task (digit, spatial), accuracy (error, correct),
target (go, nogo), and learning (pre, post). Although learning is
a likely to be a dynamic and continuous phenomenon, for
analytic purposes, it is convenient to determine a cutoff that
marks when participants have fully acquired the target–
response mapping. There are several methods for determining
when learning has occurred (Smith et al., 2004). We use the
simplest method, the fixed-number of consecutive correct
responses method1 (FCCR). In this method, a learning
1
For several subjects, we compared the fixed-number of con-
secutive correct responses method against a dynamic state–space
method for determination of a learning thershold (Smith et al.,
2004). On average, the state–space method identified the occur-
rence of learning to be two trials earlier, consistent with what
Smith et al. (2004) reported. Because analyses of the behavioral and
ERP data are based on subject averages, it is unlikely that the
reported results were affected by this small difference.
 BR A I N R ES E A RC H 1 1 7 9 ( 2 00 7 ) 8 9 –1 05
threshold was defined as the moment that participants were
able to make four consecutive correct responses (or with-
holding of a response) for a particular target. As noted in the
Experimental procedures section, the same target could not
occur on consecutive trials.
Based on the fixed-number of consecutive correct response
method, analysis of the number of trials that participants took
to learn the target–response mappings, with Target and Task
as factors, revealed a significant main effect for target, F(1,10) =
23.7, p b 0.001. Participants required more trials to learn the go
target–response mapping (M = 16.3, SD = 8.0) than withholding
a response for nogo targets (M = 8.2, SD = 2.9). This makes
intuitive sense because nogo targets do not require response
mappings. All participants reported, during task debriefing
sessions, that they focused primarily on go targets. Fig. 1 shows
accuracy (for digit and spatial tasks) as a function of trials
averaged across all stimuli.
Analysis of reaction time to targets included task,
accuracy, and learning. The analysis revealed significant
main effects for Learning, F(1,10) = 11.22, p b 0.01, and Accu-
racy, F(1,10) = 11.6, p b 0.02, that were qualified by a Learning ×
Accuracy interaction, F(1,10) = 8.3, p b 0.02. The interaction
Fig. 1 – Accuracy and Feedback processing as a function of trials for digit and spatial tasks. Vertical line marks the average trial
at which subjects reached the learning criterion.
91
revealed that participants' responses were significantly
longer for error responses than correct responses and that
this difference was greater for the post-learned state than the
pre learned state.
Because the task permitted participants to regulate the
duration of the feedback stimulus, we can use duration of
feedback processing to assess whether participants took
more time to view the feedback prior to and after learning
of the mappings. The assumption is that prior to knowing the
target–response mappings, participants will take more time
viewing the feedback. For this analysis, task and learning
factors were included. The results only showed a main effect
for Learning, F(1,10) = 11.1, p b 0.01. Participants viewed the
feedback longer prior (M = 2215 ms, SD = 1977 ms) to learning
than after learning (M = 531 ms, SD = 164 ms). Feedback
processing as a function of trials, averaged across all stimuli,
is presented Fig. 1.
The behavioral analyses demonstrate that task did not
impact the various dependent measures, which illustrate the
important fact that the tasks were equated for difficulty.
Therefore, significant differences in brain activity between
tasks cannot be attributed to task difficulty.
92 BR A IN RE S E A RCH 1 1 79 ( 20 0 7 ) 8 9 –1 05
2.2. ERP data
2.2.1. Spatiotemporal dynamics of target effects
Approximately 120 ms after target onset a clear P1 component
is maximal over occipital and occipitoparietal recording sites
(see Figs. 2 and 3). The P1 is much larger and more focal for the
digit targets compared to the spatial targets. At approximately
170 ms after target onset, the N1 reached it maximum. Like the
P1, the N1 is much larger for the digit targets. The N1 showed a
strong right lateralization for spatial targets. The amplitude
and laterality differences of these components by target type
reflect the properties of the target themselves; the digits were
presented foveally and the spatial targets are dots presented
parafoveally to demarcate spatial locations.
Interestingly, the maximum of the N1 has a different
spatial distribution than the negativity that appears to initiate
its onset. For the digit targets, the onset of the N1 began over
the left temporal recording sites, which can be seen in Fig. 2 at
the time of the P1. For the spatial targets, the onset of the N1
occurred over right temporoparietal sites and spreads caud-
ally to occipitotemporal regions.
Soon after the maximum of the N1 clear laterality
differences emerged for digit and spatial targets, starting
from occipitotemporal recording sites and spreading to
inferior temporal and frontal locations (see Fig. 4). This
negativity peaked approximately 390 ms after target onset at
frontal recording sites. For the digit targets, the negativity is
left lateralized whereas it is right lateralized for the spatial
targets. We refer to this as the lateral inferior anterior
negativity (LIAN).
2.2.2. Electrophysiological signs of learning
At approximately 327 ms after target onset a negativity can be
observed over the medial frontal channels, in particular FCz
(see Fig. 5). Based on the latency and scalp distribution of this
component, we refer to it as the MFN (see Gehring and
Willoughby, 2002). The latency and distribution of the MFN is
similar to a component often referred to as the N2 in previous
ERP studies; given the improved topography of the dense
array EEG, we prefer to emphasize the medial frontal dis-
tribution of this effect, rather than the wave deflection alone.
The MFN is most prominent for the post-learning trials for
both types of targets. At approximately the same time as the
MFN, a P300 component began to develop and peaked ap-
proximately 498 ms after target onset. The P300 is larger for
both target types after participants learned the visuomotor
association (see Fig. 6). Interestingly, the distribution of the
P300 differed according to target type. For the digit targets
(which were foveally presented), the P300 has a centroparietal
distribution whereas for spatial targets it had an occipitopar-
ietal distribution.
2.2.3. Statistical analysis
We focus on three components for statistical analysis: LIAN,
MFN, and P300. We quantified the LIAN by first identifying the
negative peak between 350 and 450 ms after target onset (see
Fig. 4). Next the average amplitude was obtained over a 44-ms
interval centered on the peak (inclusive). The data were
referenced to a baseline extending from − 400 ms to target
onset (time 0). This was done for five channels on each
hemisphere (see Fig. 7). For the left hemisphere, this included
channels F9, FT9, and three channels inferior to both these
locations, and for the right hemisphere channels F10 and FT10
and three inferior channels (homologous to the left hemi-
sphere) were included. For each hemisphere, the data were
averaged over the five channels.
The MFN was quantified at four medial frontal channels,
including FCz and Fz (see Fig. 7), by identifying the negative
peak between 260 and 350 ms after target onset (see Fig. 5)
and averaging over a 44-ms interval centered on the peak
(inclusive). Next, we identified the positivity preceding the
MFN and obtained an average over a 44-ms interval
centered on the positive peak. This served as the reference
for the MFN. The MFN was then averaged over the four
channels.
The P300 was quantified by first identifying the positive
peak between 350 and 700 ms after target onset (see Fig. 6) and
averaging over a 44-ms interval centered on the peak
(inclusive). The data were referenced to a baseline extending
from − 400 ms to target onset. This was done for individual
channels in six regions (see Fig. 7). For each region, the average
of the channels was obtained.
For analysis of the LIAN, a repeated measures ANOVA was
employed, with learning, task, and laterality (left, right) as
factors. There was a main effect for Learning, F(1,10) = 5.8,
p b 0.04, that was qualified by a significant Learning × Task
interaction, F(1,10) = 9.2, p b 0.02. This interaction indicated
that after participants learned the stimulus–response map-
pings, the LIAN was reduced in the spatial task whereas it
increased for the digit task. Mean comparisons revealed that
this difference was only significant for the spatial task,
F(1,10) = 10.6, p b 0.01. A Task × Laterality interaction was also
significant, F(1,10) = 7.6, p b 0.03. Mean comparisons confirmed
the observation that the LIAN was larger over the right
hemisphere, compared to the left hemisphere, for the spatial
task, F(1,10) = 6.4, p b 0.03. Although not significant in the
analysis, for the digit task the LIAN was larger over the left
hemisphere than the right hemisphere.
The factors learning and task were used in a repeated
measures ANOVA for the MFN component. This analysis
revealed significant Learning, F(1,10) = 15.2, p b 0.01, and Task,
F(1,10) = 18.6, p b 0.01, effects. The MFN was substantially
more negative after participants learned the task, and this
component was more negative for the verbal compared to the
spatial task.
Analysis of P300 amplitude included learning, task, lobe
(parietal, occipitoparietal), and laterality (left, midline, and
right) as factors in a repeated measures ANOVA model. The
analysis revealed a significant two-way interaction involving
task and lobe, F(1,10) = 11.0, p b 0.01, and a significant three-
way interaction involving learning, task, and laterality,
F(2,20) = 5.6, p b 0.02 (Geisser–Greenhouse corrected). For the
two-way interaction, mean comparisons revealed that P300
amplitude was largest at parietal sites, compared to occipi-
toparietal sites, for the digit task, F(1,10) = 47.1, p b 0.01, and
that this same difference was marginally significant for the
spatial task, F(1,10) = 4.7, p b 0.06.
The three-way interaction was examined by analyzing the
Learning × Laterality interaction for each task. For the spatial
task, the two-way interaction was not significant, although
 BR A I N R ES E A RC H 1 1 7 9 ( 2 00 7 ) 8 9 –1 05
Fig. 2 – Topographic waveform plots for pre- and post-learning averages for the digit and spatial tasks. Orientation of map is top looking down with nose at top of page.

93
94 BR A IN RE S E A RCH 1 1 79 ( 20 0 7 ) 8 9 –1 05

Fig. 3 – Topographic maps and waveform plots for pre learned correct go targets. Orientation of topographic maps is top
looking down with nose at top of page. Topographic maps are presented for the peaks of the P1 and N1. Dots on topographic
maps represent channel locations of the waveform plots.

Fig. 4 – 3D topographic maps and waveform plots for pre learned correct go targets. Topographic maps are presented for the
peak of the LIAN. White circles represent channel locations of the waveform plots. Large and small boxes in waveform plots
mark the time window used to quantify the P300 and LIAN, respectively.
 BR A I N R ES E A RC H 1 1 7 9 ( 2 00 7 ) 8 9 –1 05 95

Fig. 5 – Topographic maps and waveform plots for correct go targets. Topographic maps are presented for the peak of the MFN.
Dot on topographic maps represent channel location of the waveform plots. Box in waveform plot mark the time window used
to quantify the MFN.

the direction of the P300 amplitude is such that it is
consistently larger across the left, midline, and right recording
sites post-learning. For the digit task, this same interaction
was significant, F(2,20) = 4.5, p b 0.03 (Geisser–Greenhouse cor-
rected), confirming that the P300 amplitude was largest post-
learning and that this difference was largest over the left
hemisphere recording sites.
Using the Wisconsin Card Sorting task, Barceló and
colleagues (2000) found that P300 amplitude increased as
participants learned the sorting rule. To examine similar

Fig. 6 – Topographic maps and waveform plots for correct go targets. Topographic maps are presented for the peak of the P300.
Dots on topographic maps represent channel locations of the waveform plots. Large and small boxes in waveform plots mark
the time window used to quantify the P300 and LIAN, respectively.
96 BR A IN RE S E A RCH 1 1 79 ( 20 0 7 ) 8 9 –1 05
Fig. 7 – Channel groups used for quantification of the LIAN (black), MFN (red), and P300 (blue).
learning effects, we analyzed the progressive changes in each
component during the learning process. For these analyses,
we divided post-learning trials into four quartiles and
obtained an average ERP for each quartile. The components
were as previously described, and ANOVAs for each compo-
nent included the same factors, with the exception that the
learning factor now contained five levels (pre, post1, post2,
post3, post4).
The post-learning stages were defined using FCCR cri-
teria: post1 = FCCR ≥4 and ≤8; post2 = FCCR ≥9 and ≤14;
post3 = FCCR ≥ 15 and ≤22; and post4 = FCCR ≥23. This resulted
in relatively equal number of trials for each of the post-learning
stages (digit task: post1 = 45.5, post2 = 56.8, post3 = 55.2,
post4 = 60.3, spatial task: post1 = 67.3, post2 = 65.8, post3 = 56.8,
post4 = 74.2). For each component, we focus on the learning
effects that were found to be significant in the preceding
analyses in order to clarify the nature of the changes during the
learning process.
For the LIAN, the Learning × Task interaction was signifi-
cant, F(4,40) = 4.57, p b 0.02 (Geisser–Greenhouse corrected).
Analysis of the main effects revealed that the learning effect
was only significant for the spatial task, F(4,40) = 3.75, p b 0.03
(Geisser–Greenhouse corrected), which is similar to the
analyses reported above. This main effect for the spatial task
revealed that LIAN amplitude continued to decrease as
learning progressed (see Fig. 8).
Analysis of the MFN revealed a significant learning effect
(see Fig. 8), F(4,40) = 8.16, p b 0.01 (Geisser–Greenhouse cor-
rected). This effect illustrates that MFN amplitude continued
to increase as learning progressed. Similar to the MFN re-
sults, analysis of the P300 revealed a significant main effect
for Learning, F(4,40) = 10.2, p b 0.01, that did not significantly
interact with any other factor. This effect showed that P300
amplitude increased as participants learned the task (see
Fig. 8).
It should be noted that although there is temporal overlap
between the LIAN, MFN, and P300, our quantification of these
components rely on temporal integration around a specific
peak to minimize the overlap. It is likely that the component
quantification is imprecise with regards to component over-
lap, particularly for the LIAN and P300. To have a better sense
of the uniqueness of each component in relation to learning,
we supplemented the analysis by employing principal com-
ponents analysis (PCA).
2.2.4. Spatial–temporal principal components analysis
We submitted the averaged data to PCA using ERP PCA Toolbox
(http://people.ku.edu/~jdien/downloads.html, see Dien et al.,
2003). For this analysis, a covariance matrix of the channels
was employed. The results were rotated into simple structure
using the Promax rotation procedure. From this analysis, we
kept four components that accounted for 96% of the variance.
The components are shown in Fig. 9. Components 1 and 4
reflect the LIAN, component 2 resembles the P300, and
component 3 clearly represents the MFN. Fig. 9 also shows
the respective factor scores for each component.
 BR A I N R ES E A RC H 1 1 7 9 ( 2 00 7 ) 8 9 –1 05
Fig. 8 – (A) LIAN amplitude as a function of learning. (B) MFN and P300 amplitudes as a function of learning.
The factor scores of the four components were then
submitted to a temporal PCA using a covariance matrix of
time points. The results were rotated into simple structure
using the Promax rotation procedure. For each spatial
component, we retained 15 temporal components and identi-
fied the relevant component(s) for each spatial component
based on the time window and condition differences specified
from the scalp data. Fig. 9 shows the relevant temporal
components. For spatial components 1, 2, 3, and 4, the
corresponding identified temporal components accounted
for 12%, 21%, 2%, and 6% of the variance.
Repeated measures ANOVAs were performed on the
identified temporal components. For statistical analysis, we
employed the same time windows used for analysis of the
scalp data. For example, the time window of the LIAN was
employed for the temporal components of components 1
and 4. For analysis of the LIAN equivalent, learning, task,
and component (1 and 4) served as repeated measures
factors. The results are identical to results obtained with the
scalp data; there were significant interactions involving
Learning and Task, F(1,10) = 6.19, p b 0.04, and Task and
Component, F(1,10) = 11.66, p b 0.01. For the Learning × Task
97
interaction, a means comparison confirmed that the LIAN
decreased as subjects learned the spatial task, F(1,10) = 10.5,
p b 0.01. The difference between pre and post-learning for
the verbal task was not significant. A means comparison of
the Task × Component interaction reveal that component 4
was larger for the spatial than the digit task, F(1,10) = 10.5,
p b 0.01, see Fig. 10. Although component 1 was larger for
the digit task than the spatial task, this difference did not
reach statistical significance.
Analysis of component 2 included learning and task as
repeated measures factor. There was a significant learning
effect, F(1,10) = 30.3, p b 0.01, showing that the component was
largest after learning. For component 3, learning and task also
served as repeated measures factors. The analysis revealed
significant main effect for Learning, F(1,10) = 12.71, p b 0.01, and
Learning × Task interaction, F(1,10) = 7.08, p b 0.03. This interac-
tion revealed that the amplitude increase in the post-learning
state was largest for the digit task.
These analyses are not complicated by potential compo-
nent overlap, and the results confirm the results obtained
from the scalp data. Therefore, it is significant that learning
effects were observed for the LIAN, MFN, and P300.
98 BR A IN RE S E A RCH 1 1 79 ( 20 0 7 ) 8 9 –1 05

Fig. 9 – Topographic maps of the spatial PCA component loadings, spatial factor scores, and temporal factor loadings. Box in
waveform plot mark the time window used to quantify each component. Boxes correspond to those presented in Figs. 4–6.

2.2.5. Source estimates
Source estimates of the scalp potential were accomplished
using the GeoSource electrical source imaging software (EGI,
Eugene, OR). GeoSource uses a finite difference model (FDM)
for accurate computation of the lead field in relation to cranial
orifices (primarily optical canals and foramen magnum).
Tissue compartments of the model were constructed from
whole-head MRI and CT scans of a single subject.2 The MRI
and CT images were coregistered prior to segmentation of the
brain and cerebral spinal fluid, derived from MRI data, and the
skull and scalp, identified from CT images. The tissue volumes
were parceled using 2-mm voxels to form the computational
elements of the FDM.
Conductivity values used in the FDM model are as follows:
0.25 S/m (Siemens/meter) for brain, 1.8 S/m for cerebral spinal

Fig. 10 – Task × component interaction.
2
This subject is the same subject whose average MRI is used in
SPM99 (http://www.fil.ion.ucl.ac.uk/spm). Because the average
MRI of this subject does not contain data from the inferior surface
of the head and the skull cannot be easily extracted from MRI
data, EGI, in collaboration with this subject, acquired whole head
MRI and CT scans of this subject to construct a whole-head
model.
 BR A I N R ES E A RC H 1 1 7 9 ( 2 00 7 ) 8 9 –1 05 99

fluid, 0.018 S/m for skull, and 0.44 S/m for scalp (see Ferree et
al., 2000). Source locations were derived from the Montreal
Neurological Institute probabilistic MRI (to which the typical
subject matches closely). Based on the probabilistic map, gray
matter volume was parceled into 7 mm voxels, each voxel
served as a source location with three orthogonal orientations.
This resulted in a total of 2,394 sources. The source locations
were registered to the MRI and CT volumes. Once the head
model was constructed, an average of the 128-channel
positions was registered to the scalp surface. To compute
estimates of the sources, a minimum norm solution with the
LAURA (local autoregressive average) constraint (Grave de
Peralta Menendez et al., 2004) was employed.
The estimated source regions contributing to the LIAN, at
its peak, are illustrated in Fig. 11A. For the digit task, the LIAN
is associated with left lateralized activity in the medial
temporal lobe (including the parahippocampal gyrus), lateral
temporal lobe, and inferior frontal gyrus (BA 47). For the
spatial task, the sources for the LIAN are in the right medial
temporal lobe (including the parahippocampal gyrus), anterior

Fig. 11 – (A) Source estimates for LIAN component for digit and spatial targets. Lines at each voxel represent orientation vectors
(pointing in the positive direction). The vectors indicate the scalp topography features that are accounted by the source voxels.
Shown are sources in the temporal and ventrolateral (top row) and parahippocampal regions (bottom row). (B) Source estimates
for MFN component for digit and spatial targets in the post-learned state. Note that other sources not included in the clusters,
but illustrated, account for other aspects of the scalp topography that are not the MFN (see Fig. 3). Brighter colors denote stronger
current sources.
100 BR A IN RE S E A RCH 1 1 79 ( 20 0 7 ) 8 9 –1 05

Fig. 12 – Source estimates for P300 component for digit and spatial targets. Top row shows sources in medial and lateral
parietal cortex, BA 7, and BA 7/BA 40, respectively, and parahippocampal gyri. Note laterality differences in the
parahippocampal sources as a function of target type. Brighter colors denote stronger current sources.

temporal cortex (including superior and inferior temporal
gyri), and inferior frontal gyrus (BA 47).
For the MFN associated with the digit task in the post-
learned state, sources (see Fig. 11B) were identified in three
clusters: (1) medial frontal gyrus (BA 6 and 8), (2) dorsal ACC (BA
24 and 32) and paracentral lobule (BA 31), and (3) ventral medial
frontal gyrus (BA 10), orbital gyrus (BA 11), and ventral ACC (BA
32). Similar sources of the MFN were identified for the spatial
task in the post-learned state, with the exception of cluster 1.
The P300 sources for the digit task (estimated from the post-
learning data, see Fig. 12) are localizable to the medial (BA 7)
and lateral aspects of the left parietal lobe (BA 7, 40), right
superior temporal gyrus (BA 39), posterior cingulate cortex (BA
30, 31), medial occipital cortex (BA 17, 18, and 19), and bilateral
parahippocampal gyrus (BA 36 and 37). The pattern was similar
for the P300 for the spatial task. In addition, however, there
were sources in the right parietal lobe (BA 7 and 40) and medial
frontal gyrus (BA 6). It is noteworthy that the major sources for
the P300, in parahippocampal regions, demonstrated greater
activity in the right compared to the left hemisphere for the
spatial task, and this was reversed in the digit task.
3. Discussion
Results of this study provided clues to progressive engage-
ment of brain networks in the process of learning. In early
learning stages, when participants are still trying to associate
visual information with appropriate responses, lateralized
negativities originated with the onset of the N1. These
lateralized negativities (left for digit code, right for spatial
code) spread from lateral temporal recording sites to inferior
anterior sites, peaking at approximately 390 ms post-target.
We measured the peak of this effect as the LIAN.
After participants learned the task, there were three
prominent changes identified in the ERPs. The first was a
reduction in LIAN amplitude, particularly for spatial targets.
This seems consistent with the hypothesis of a decrease in
frontal contributions to controlled processes as learning
progresses. The other two changes involved increases in
MFN and P300 amplitude in the post-learned state. Whereas
the increase in P300 was predicted and is consistent with
previous research, the increase in the MFN with learning was
unexpected. We consider each of these components in turn.
3.1. LIAN
LIAN amplitude showed some signs of tracking the learning
process, but differently for the two tasks. For spatial targets,
after participants learned the stimulus–response mappings,
LIAN amplitude was significantly reduced as was hypothe-
sized. However, although the effect was not significant, LIAN
amplitude increased after learning the correct responses for
the digit targets. This increase could be due to component
 BR A I N R ES E A RC H 1 1 7 9 ( 2 00 7 ) 8 9 –1 05
overlap because the increase of the LIAN may actually reflect
P300 amplitude increase (the P300 is positive at parietal, but
negative at inferior frontal and temporal regions). However,
analysis of the PCA components (1 and 4), which addresses the
overlap problem, revealed similar results as analysis of the
scalp ERPs, confirming the differential decrease of the LIAN by
task. There are two issues related to the functional interpreta-
tion of these findings: the functional significance of the LIAN
and the differential rate of learning-related reduction.
Source estimates for digit targets revealed that during the
time of the LIAN, activity was lateralized to left temporal lobe
and left inferior frontal gyrus. For spatial targets, similar
sources were identified in the right hemisphere. Studies using
fMRI methods have shown activations in these same cortical
regions during arbitrary visuomotor (Grol et al., 2006; Toni and
Passingham, 1999; Toni et al., 2001) and sequence learning
(Poldrack et al., 2005; Van der Graaf et al., 2006). In primates,
ventral and orbital regions of the prefrontal cortex have been
shown to be involved in the early learning stages of arbitrary
visuomotor mapping, perhaps through the acquisition of
response strategies (Bussey et al., 2001). In humans, the role
of frontal and temporal regions in learning has been attributed
to memory operations, in which targets take on associative
meaning (Bunge et al., 2003; Toni et al., 2001). Results from
several studies have suggested that ventrolateral prefrontal
cortex is involved in memory retrieval through its interaction
with temporal lobe structures, including the hippocampus
(Badre et al., 2005; Simons and Spiers, 2003).
In a meta-analytic analysis of 29 studies, Chein and
Schneider (2005) reported that 8 of 12 brain regions that
showed practice related decreases were in the right hemi-
sphere. Of particular relevance to the present LIAN results is
the finding of right lateralization of practice-related decreases
in the insula and surrounding cortical regions (including the
ventral aspects of the frontal lobe and anterior temporal lobe).
Similarly, a recent study using target locations to guide
sequence learning demonstrated right lateralized involve-
ment of the ventrolateral prefrontal cortex early in learning
(Van der Graaf et al., 2006).
A relevant recent finding is that visual orienting (used for
selecting stimuli for action) is anatomically and functionally
separable into two sub-systems: a dorsal system (bilateral
intraparietal sulcus and bilateral superior frontal sulcus) and a
ventral system (right temporal-parietal junction and right
ventral frontal cortex; Fox et al., 2006). The ventral system is
believed to be involved in detection of salient stimuli.
Functionally, early in learning it is likely that to-be-learned
stimuli are particularly salient because participants must map
each stimulus to an appropriate response. Thus, based on
previous research and the characteristics of the present tasks,
we might believe that activity of the right hemisphere
(particularly the ventrolateral prefrontal structures) could
decrease at a faster rate as a function of learning than the
activity in homologous networks the left hemisphere.
It is important to note that the right ventrofrontal and
anterior temporal activation indexed by the LIAN is present
during learning of both the digit and spatial tasks. Therefore,
this system must be critical for learning generally, although
the decrease in right ventral frontal activity was only observed
for the spatial task. Our speculative interpretation at this point
101
is that the right ventral frontal activity was particularly
important for attentional orienting to the configural features
of the spatial task, and thus reflected hemisphere-specific
processing that declined as learning increased within the right
hemisphere specifically.
3.2. MFN
In the present study, MFN amplitude associated with the
target increased as participants learned the task. This effect
was opposite to the hypothesis that frontal contributions
would decrease with increasing expertise and appears to be a
novel finding. For the response-locked medial frontal compo-
nent, the error-related negativity (ERN), previous research has
shown this component did not appear until subjects had
learned the task (Dehaene et al., 1994). The increase in MFN
amplitude with increasing learning in the present study may
therefore reflect an improving model of the performance
context that is used to guide performance (Luu and Tucker,
2003). Previous research with the medial frontal negativity has
shown it to be sensitive to evaluation of negative feedback
(Luu and Tucker, 2003; Tucker et al., 1999) and to perception of
reward reduction or loss (Gehring and Willoughby, 2002;
Yeung et al., 2005). These studies have examined the MFN in
response to feedback, not the target stimulus. In these cases,
the feedback-locked MFN decreases with learning (see Holroyd
and Coles, 2002).
To examine the similarity between the MFN of the present
study with previous dense array MFN findings, using the
improved source localization with the FDM head model in
GeoSource, we compared the present source distribution of
the MFN with that from our previously published data (Luu et
al., 2003). The source estimate, using the same parameters for
GeoSource analysis described previously, showed a source in
dorsomedial frontal cortex for the Luu et al. (2003) results that
is almost identical to the dorsomedial frontal source for the
present MFN (cluster 2 in Fig. 11). Notably different for the
present data, however, were additional sources, particularly in
the rostroventral prefrontal cortex (cluster 3), that were also
identified as contributing to the MFN. With the appearance of
activity in this rostroventral midline region, the source model
for the present MFN more closely resembles that for our
previous ERN results (Luu et al., 2003) rather than the previous
MFN results.
The MFN response in the present experiment can be
integrated with previous results by adopting a learning
model of regulating actions in context (Luu et al., 2003). This
model is derived from Gabriel's theoretical analysis of cortico-
thalamic-limbic circuits in discriminant avoidance learning
(Gabriel, 1993). In this analysis, the amygdala and ACC form a
system involved in the early stages of learning to discriminate
between conditioned stimuli that predict delivery of a shock
versus those that do not (Gabriel et al., 2002). Cells within this
ACC circuit become activated in a discriminant manner as the
animal acquires the differentiating behavior for a new context.
Consistent with this analysis, a recent visuomotor association
learning study found cells within the medial prefrontal cortex
that differentiated between novel and familiar cues, with
neurons sensitive to familiar cues more active when new
learning was required (Inase et al., 2006). Single cell recordings
102 BR A IN RE S E A RCH 1 1 79 ( 20 0 7 ) 8 9 –1 05
in the human ACC also show cells that are sensitive to reward
reductions that signal requirement for behavior change
(Williams et al., 2004). These several findings suggest that
the ACC supports rapid motivated learning, particularly when
actions must be regulated strongly and quickly to adapt to new
contextual contingencies. Therefore, one way to interpret the
increasing MFN amplitude with learning is to consider this
component as reflecting monitoring functions that are
required as subjects learn appropriate actions for a given
context. Once the context (i.e., stimulus–response mappings)
is acquired in the present task, potential for response conflict
becomes established more strongly. In fact, the behavioral
data show that errors after learning are associated with the
longest reaction times, consistent with a view of increased
conflict after learning because the correct response is now
established. In this sense, therefore, the MFN may reflect
mechanisms related to conflict monitoring functions of the
ACC (Botvinick et al., 2004). Based on the interpretation that
the MFN reflects response and conflict monitoring that
increase as the correct actions and context are learned, we
predict that with additional practice, where stimulus–
response mappings are further automated, involvement of
the ACC (as indexed by the MFN) would decrease, consistent
with prior studies of ACC and automaticity (Raichle et al.,
1994).
3.3. P300
In studies using both fMRI and ERP methods, sources of the
scalp recorded EEG have been identified in the parietal cortex
(BA 40; Menon et al., 1997), bilateral posterior parietal cortex,
inferior temporal cortex, temporoparietal junction, and ven-
trolateral prefrontal cortex (Bledowski et al., 2006). In the
present study, we found roughly similar sources in the parietal
cortex (both medial and lateral locations, BA 7). Additionally,
we found sources in posterior cingulate cortex, medial
occipital cortex, superior temporal cortex, with the prominent
source moments in bilateral parahippocampal regions. These
sources were common to the P300 for both digit and spatial
target types. For spatial targets, there were additional sources
in right parietal and medial frontal cortices (BA 6).
Research using fMRI methods have identified increases in
lingual gyrus (BA 19) and parahippocampal gyrus during
acquisition of arbitrary visuomotor mappings (Toni and
Passingham, 1999). Recently, Grol and colleagues (2006)
found increases in activity in three regions when participants
overlearn visuomotor mappings: intraparietal sulcus, medial
BA 6, and BA 45/46. These brain regions are similar to the P300
sources we identified in the present study. Grol et al. noted
that activity in parietal cortex correlated with the degree of
automaticity that participants exhibited. These authors sug-
gested that learning related changes observed in the parietal
cortex might reflect consolidation of procedural memories,
perhaps through slow Hebbian-like plasticity changes. These
findings may be consistent with the notion that automaticity
in learning continues to engage activity in parietal regions
(Chein and Schneider, 2005).
In the present study, as participants gained increasing
skill with the code learning tasks, they showed a correspond-
ing increase in the amplitude of the P300. Learning studies
using ERP measures have consistently shown increases in
P300 amplitude after learning. Using the Wisconsin card
sorting task these, Barceló and colleagues (2000) discovered
that P300 amplitude was reduced whenever participants
were required to learn a new rule, and it gradually increased
as a new rule was acquired. This led to the hypothesis that
the P300 reflects cortical activity associated with stable
representation of rule-coding (or attentional) sets. We think
this is consistent with the classical notion of context-
updating (Donchin and Coles, 1988), and with our interpreta-
tion that context-updating is supported by the dorsal limbic
(Papez circuit) late stage learning system (Luu and Tucker,
2003).
3.4. Source estimation issues
The analysis of neural mechanisms of scalp EEG can be greatly
facilitated by source analysis. In describing the present results,
we have called out the specific gyri and regions implied by the
atlas based cortically constrained source model of the
GeoSource software. However, with only 128 scalp channels,
without exact registration of sensor positions, and with the
limitation of the atlas reconstruction rather than the subject's
own cortex, we estimate that a 1-cm3 error is the minimum to
be expected. Therefore, we can be fairly confident that the
P300, for example, is localized to medial temporal regions, but
any further precision is probably spurious. It should be
emphasized that the present 128 channel sensor array (the
HydroCel Geodesic Sensor Net) was constructed to provide
improved coverage of the inferior head regions (face and neck)
compared to the previous 128 array (GSN 200). The present
results show that, regardless of source localization, without
adequate coverage of inferior head regions, components such
as the LIAN or P300 cannot be measured in the full extent of
their volume conduction through head tissues.
3.5. Conclusion
Learning is a dynamic and continuous process, wherein
expertise and automaticity may be the end state. The present
results do not address potential changes associated with
extended practice. Rather, the results reveal that the transi-
tion between not knowing what to do and performing
appropriate actions in context (learning) is marked by
differential decreases in right frontal lobe activation and
increased activation in the ACC and posterior networks of
the temporal lobe, PCC, and parietal cortex.
Unlike fMRI studies, EEG and ERP methodologies show the
time course of brain activity. The results revealed substantial
temporal overlap between processes indexed by three ERP
components. The LIAN is initiated quite early and appears to
be a continuation of the neural activity that extends along the
extent of the temporal lobe, perhaps reflecting memory
operations in occipitotemporal networks. The strong activa-
tion of inferior frontal and anterior temporal regions (at the
peak of the LIAN) overlaps temporally with ACC (MFN), PCC,
medial temporal, and parietal activations. The data also
suggest that the course of learning related to the processes
regulated by these structures is quite different. Activity
recorded from frontal and temporal structures, at least the
 BR A I N R ES E A RC H 1 1 7 9 ( 2 00 7 ) 8 9 –1 05
right-hemisphere response associated with spatial targets,
decrease at a faster rate than activity in the ACC. It remains to
be seen whether left-hemisphere activity (indexed by the
LIAN) and ACC activity associated with conflict monitoring
would decrease, as expected, with extended practice.
These electrophysiological signs of early learning pro-
cesses are followed by a gradual development of the P300,
reflecting activation within medial and posterior networks of
the temporal lobe, as well as associated networks of the
posterior cingulate and parietal cortices. The involvement of
these posterior cortical structures mark the transition to the
later stage of learning and perhaps the onset of automaticity.
4. Experimental procedures
4.1. Participants
Eleven undergraduates students were recruited from the
University of Oregon Psychology Department subject pool.
Prior to participation in the study, all participants gave inform
consent. The study was approved by EGI and the University of
Oregon's institutional review boards. Participants were paid
$15 for their participation and an additional amount, ranging
between $25 and $45, depending on task performance. On
average participants earned $40. All participants had normal
or corrected-to-normal vision and none reported taking
medications or drugs. The participants were all right handed
and five were men. Their age ranged between 19 and 32
(M = 24.54, SD = 4.5).
Prior to application of the 128-channel HydroCel Geodesic
Sensor Net (Electrical Geodesics, Inc., Eugene, OR), participants
completed several brief mood questionnaires. After prepara-
tion for EEG recording, participants were seated 55 cm in front
of a computer monitor. Distance from the monitor and
alignment were controlled by a fixed chin rest. The chin rest
also helped to minimize head movements.
Fig. 13 – Diagram of events within a single trial for the digit and spatial tasks.
103
4.2. Tasks
Each participant completed two studies, the order of which
was counterbalanced. The time lapse between studies was not
greater than seven days. Both studies were identical with
regards to event timeline and stimulus duration (see Fig. 13).
Each trial began with a presentation of a target (1500 ms max
duration). Targets were randomly presented, with the con-
straint that the same target could not occur on consecutive
trials. The target was terminated when participants made a
response or 1500 ms elapsed. After target termination, a
feedback stimulus was presented to inform participants of the
status of their response. There were six possible forms of
feedback: (1) ErrorGo (error of omission in response to go
target), (2) ErrorNG (error of commission in response to nogo
target), (3) Correct (correct response to go target but response
committed with wrong hand), (4) CorrectH (correct response
with correct hand to go target but response committed with
wrong finger), (5) CorrectNG (correct withholding of response
to nogo target), and (6) CorrectF (correct response with correct
hand and finger to go target). The feedback was presented for a
maximum duration of 10 s, unless terminated by the subject
with a button press. The inter-trial interval varied between
1500 and 2500 ms. There was a total of 800 trials for each task
and the trails were grouped into 100 trial blocks.
Participants were informed that correct feedback (Cor-
rectNG and CorrectF) results in 8 points earned, Error feedback
(ErrorGo and ErrorNG) results in 8 points lost, and partial
correct feedback, Correct and CorrectH, result in loss of 4 and 2
points, respectively. Participants started the study with zero
points. To motivate participants to learn the task, participants
were informed that each feedback also denote points earned
or lost and that their goal should be to earn as many points as
possible because they will be paid a monetary bonus according
to the amount of points they accumulate by the end of the
study. In between each block, participants took a brief break
and recorded their accumulated points on a form.
104 BR A IN RE S E A RCH 1 1 79 ( 20 0 7 ) 8 9 –1 05
4.2.1. Digit task
In the digit task, target stimuli were two-digit numbers
(courier new bold font type, 18 font size). A target replaced
the fixation mark at the start of each trial. Participants were
informed that during the course of the study they will be
presented with a total of 16 two-digit numbers and that their
task is to learn, through trial and error, which numbers
required a response and which numbers required no response.
For those that required a response, they were instructed to
learn which hand and finger they should use to press a button.
Eight of the two-digit numbers were go targets. Participants
responded with the middle and index finger of each hand (two
go targets were mapped onto each finger).
Prior to the actual task, participants were provided with 32
practice trials (16 go trials). In the practice trials, only four (two
go and two nogo targets) two-digit numbers were presented
and these were mapped onto left or right hand only (i.e.,
participants did not have to determine the finger mapping).
The practice targets are unique in that they are not repeated
during the actual study. The practice trial was to orient
participants to the trial-and-error learning nature of the task.
All participants were able to learn the response mapping by
the end of the practice session.
4.2.2. Spatial task
Spatial locations were created by spacing a dot every 22.5°
around a fixation mark (radius of ∼ 4 cm) for a total of 16
locations. Based on participants distance from the monitor,
the dots were presented approximately 4.2° from fixation.
Participants were informed that during the course of the
study, a dot will be presented in eight different locations3 (four
go and four nogo locations randomly selected from 16
locations) around fixation and that their task is to learn,
through trial and error, which location required a response
and which location required no response. For those that
required a go response, they were instructed to learn which
hand and finger they should use to press a button. Participants
responded with the middle and index finger of each hand (one
go target was mapped onto each finger).
Prior to the actual task, participants were provided with 32
practice trials (16 go trials). In the practice trials, four (two go
and two nogo targets) target locations were presented and
these were mapped onto left or right hand only (i.e.,
participants did not have to determine the finger mapping).
The practice targets are unique in that they are not repeated
during the actual study. All participants were able to learn the
response mapping by the end of the practice session.
4.3. EEG recording
The EEG was acquired using a 128-channel HydroCel Geodesic
Sensor Net (Electrical Geodesics, Inc., Eugene, OR). All electro-
des were kept below 70 KΩ (Ferree et al., 2001). All recordings
were referenced to Cz. The EEG was bandpass filtered (0.1–
100 Hz) and sampled with a 16-bit analog-to-digital converter
3
In pilot studies, it was determined that participants found the
task to be very difficult when 16 target positions (all positions)
were employed. To equate task difficulty with the digit task, we
had to reduce the target locations to eight.
at 250 s/s. Blink-contaminated EEG intervals were corrected in
the continuous record prior to segmentation of the data (Ille
et al., 2002).
The data were then segmented relative to target onset
(400 ms before and 1000 ms after) and sorted according to pre
and post-learning (see Results section). Each segment of the
EEG was excluded from signal averaging if 10 or more
channels contain data that exceed a voltage threshold of
200 μV (absolute) or a transition threshold of 100 μV (sample
to sample). After averaging, the data were re-referenced to
the average reference. Only go targets are examined because
we are specifically interested in elucidating the neural
mechanisms responsible for visuomotor mapping. Errors
were not analyzed because not all subjects had enough pre-
and post-learning trials for ERP averages. Overall, all error
types made up 41% of trials for the digit task and 33% for the
spatial task.
Acknowledgments
This project was supported by the Office of Naval Research,
“Cognitive Performance Enhancement” and VIRTE pro-
grams, DARPA contract W31P4Q-04-C-R250, and NIMH grant
MH070911.
REFERENCES
Badre, D., Poldrack, R.A., Paré-Blagoev, J., Insler, R.Z., Wagner, A.D.,
2005. Dissociable controlled retrieval and generalized selection
mechanisms in ventrolateral prefrontal cortex. Neuron 47,
907–918.
Barceló, F., Muñoz-Céspedes, J.M., Pozo, M.A., Rubia, F.J., 2000.
Attentional set shifting modulates the target p3b response in
the Wisconsin card sorting test. Neuropsychologia 38,
1342–1355.
Bledowski, C., Kadosh, K.C., Wibral, M., Rahm, B., Bittner, R.A.,
Hoechstetter, K., et al., 2006. Mental chronometry of working
memory retrieval: a combined functional magnetic resonance
imaging and event-related potentials approach. J. Neurosci. 26,
821–829.
Botvinick, M.W., Cohen, J.D., Carter, C.S., 2004. Conflict monitoring
and anterior cingulate cortex: an update. Trends Cogn. Sci. 8,
539–546.
Bunge, S.A., Kahn, I., Wallis, J.D., Miller, E.K., Wagner, A.D., 2003.
Neural circuits subserving the retrieval and maintenance of
abstract rules. J. Neurophysiol. 90, 3419–3428.
Bussey, T.J., Muir, J.L., Everitt, B.J., Robbins, T.W., 1996. Dissociable
effects of anterior and posterior cingulate cortex lesions on the
acquisition of a conditional visual discrimination: facilitation
of early learning vs. Impairment of late learning. Behav. Brain
Res. 8245–8256.
Bussey, T.J., Wise, S.P., Murray, E.A., 2001. The role of ventral and
orbital prefrontal cortex in conditional visuomotor learning
and strategy use in rhesus monkeys (Macaca mulatta). Behav.
Neurosci. 115, 971–982.
Chein, J.M., Schneider, W., 2005. Neuroimaging studies of
practice-related change: fMRI and meta-analytic evidence of a
domain general control network for learning. Cogn. Brain Res.
25, 607–623.
Dehaene, S., Posner, M.I., Tucker, D.M., 1994. Localization of a
neural system for error detection and compensation. Psychol.
Sci. 5, 303–305.
 BR A I N R ES E A RC H 1 1 7 9 ( 2 00 7 ) 8 9 –1 05
Dien, J., Spencer, K.M., Donchin, E., 2003. Localization of the
event-related potential novelty response as defined by
principal components analysis. Cogn. Brain Res. 17, 637–650.
Donchin, E., Coles, M.G.H., 1988. Is the p300 component a
manifestation of context updating? Behav. Brain Sci. 11,
357–374.
Ferree, T., Eriksen, K.J., Tucker, D.M., 2000. Region head tissue
conductivity estimation for improved EEG analysis. IEEE Trans.
Biomed. Eng. 47, 1584–1592.
Ferree, T., Luu, P., Russel, J.S., Tucker, D.M., 2001. Scalp electrode
impedance, infection risk, and EEG data quality. Clin.
Neurophysiol. 112, 536–544.
Fitts, P.M., 1964. Perceptual–motor skill learning. In: Melton, A.W.
(Ed.), Categories of Human Learning. Academic Press,
New York, pp. 243–285.
Fox, M.D., Corbetta, M., Snyder, A.Z., Vincent, J.L., Raichle, M.E.,
2006. Spontaneous neuronal activity distinguishes human
dorsal and ventral attention systems. Proc. Natl. Acad. Sci. 103,
10046–10051.
Gabriel, M., 1993. Discriminative avoidance learning: a model
system. In: Vogt, B.A., Gabriel, M. (Eds.), Neurobiology of the
Cingulate Cortex and Limbic Thalamus. Birkhäuser, Boston,
pp. 478–523.
Gabriel, M., Burhans, L., Talk, A., Scalf, P., 2002. Cingulate cortex.
In: Ramachandran, V.S. (Ed.), Encyclopedia of the Human Brain.
Elsevier Science, pp. 775–791.
Gehring, W.J., Willoughby, A.R., 2002. The medial frontal cortex
and the rapid processing of monetary gains and losses. Science
295, 2279–2282.
Grave de Peralta Menendez, R., Murray, M.M., Michel, C.M.,
Martuzzi, R., Gonzalez Andino, S.L., 2004. Electrical
neuroimaging based on biophysical constraints. NeuroImage
21, 527–539.
Grol, M.J., de Lange, F.P., Verstraten, F.A.J., Passingham, R.E., Toni,
I., 2006. Cerebral changes during performance of overlearned
arbitrary visuomotor associations. J. Neurosci. 26, 117–125.
Holroyd, C.B., Coles, M.G.H., 2002. The basis of human error
processing: reinforcement learning, dopamine, and the
error-related negativity. Psychol. Rev. 109, 679–709.
Ille, N., Berg, P., Scherg, M., 2002. Artifact correction of the ongoing
EEG using spatial filters based on artifact and brain signal
topographies. J. Clin. Neurophysiol. 19, 113–124.
Inase, M., Li, B.-M., Takashima, I., Iijima, T., 2006. Cue familiarity is
represented in monkey medial prefrontal cortex during
visuomotor association learning. Exp. Brain Res. 168,
281–286.
Keng, E., Gabriel, M., 1998. Hippocampal modulation of
cingulo-thalamic neuronal activity and discriminative
avoidance learning in rabbits. Hippocampus 8, 491–510.
Luu, P., Tucker, D.M., 2003. Self-regulation and the executive
functions: electrophysiological clues. In: Zani, A., Preverbio,
A.M. (Eds.), The Cognitive Electrophysiology of Mind and
Brain. Academic Press, San Diego, pp. 199–223.
Luu, P., Tucker, D.M., Derryberry, D., Reed, M., Poulsen, C., 2003.
Electrophysiological responses to errors and feedback in the
process of action regulation. Psychol. Sci. 14, 47–53.
Menon, V., Ford, J.M., Lim, K.O., Glover, G.H., Pfefferbaum, A., 1997.
Combined event-related fMRI and EEG evidence for
105
temporal–parietal cortex activation during target detection.
NeuroReport 8, 3029–3037.
Miller, E.K., Cohen, J.D., 2001. An integrative theory of prefrontal
cortex function. Annu. Rev. Neurosci. 24, 167–202.
Poldrack, R.A., Sabb, F.W., Foerde, K., Tom, S.M., Asarnow, R.F.,
Bookheimer, S.Y., et al., 2005. The neural correlates of motor
skill automaticity. J. Neurosci. 25, 536–5364.
Posner, M.I., DiGirolamo, G., 1998. Executive attention: conflict,
target detection and cognitive control. In: Parasuraman, R.
(Ed.), The Attentive Brain. MIT Press, Cambridge, pp. 401–423.
Raichle, M.E., Fiez, J.A., Videen, T.O., MacLeod, A.-M.K., Pardo, J.V.,
Fox, P.T., et al., 1994. Practice-related changes in human brain
functional anatomy during nonmotor learning. Cereb. Cortex 4,
8–26.
Rescorla, R.A., Wagner, A.R., 1972. A theory of pavlovian
conditioning: variations in the effectiveness of reinforcement
and nonreinforcement. In: Black, A.H., Prokasy, W.F. (Eds.),
Classical Conditioning II: Current Research and Theory.
Appleton-Century-Corfts, New York, pp. 64–99.
Schneider, W., Shiffrin, R.M., 1977. Controlled and automatic
human information processing: I. Detection, search, and
attention. Psychol. Rev. 84, 1–66.
Shiffrin, R.M., Schneider, W., 1977. Controlled and automatic
human information processing: II. Perceptual learning,
automatic attending and a general theory. Psychol. Rev. 84,
127–190.
Simons, J.S., Spiers, H.J., 2003. Prefrontal and medial temporal lobe
interactions in long-term memory. Nat. Rev., Neurosci. 4,
637–648.
Smith, A.C., Frank, L.M., Wirth, S., Yanike, M., Hu, D., Kubota, Y.,
et al., 2004. Dynamic analysis of learning in behavioral
experiments. J. Neurosci. 24, 447–461.
Toni, I., Passingham, R.E., 1999. Prefrontal-basal ganglia pathways
are involved in the learning of arbitrary visuomotor
associations: a pet study. Exp. Brain Res. 127, 19–32.
Toni, I., Ramnani, N., Josephs, O., Ashburner, J., Passingham, R.E.,
2001. Learning arbitrary visuomotor associations: temporal
dynamic of brain activity. NeuroImage 14, 1048–1057.
Tucker, D.M., Luu, P., 2006. Adaptive binding. In: Zimmer, H.,
Mecklinger, A., Lindenberger, U. (Eds.), Binding in Human
Memory: A Neurocognitive Approach. Oxford University Press,
New York, pp. 85–113.
Tucker, D.M., Hartry-Speiser, A., McDougal, L., Luu, P., deGrandpre,
D., 1999. Mood and spatial memory: emotion and the right
hemisphere contribution to spatial cognition. Biol. Psychol. 50,
103–125.
van der Graaf, F.H.C.E., Maguire, R.P., Leenders, K.L., de Jong, B.M.,
2006. Cerebral activation related to implicit sequence learning
in a double serial reaction time task. Brain Res. 1081, 179–190.
Williams, Z.M., Bush, G., Rauch, S.L., Cosgrove, G.R., Eskander, E.N.,
2004. Human anterior cingulate neurons and the integration of
monetary reward with motor responses. Nat. Neurosci. 7,
1370–1375.
Wise, S.P., Murray, E.A., 2000. Arbitrary associations between
antecedents and actions. Trends Neurosci. 23, 271–276.
Yeung, N., Holroyd, C.B., Cohen, J.D., 2005. ERP correlates of
feedback and reward processing in the presence and absence
of response choice. Cereb. Cortex 15, 535–544.