Brain potentials elicited by correct and incorrect keypresses of expert pianists differed already 100 ms before keypresses were fully executed, and thus prior to the onsets of erroneous tones (pre-error negativity). The early detection of errors is also observable at the behavioral level: IOIs before erroneous keypresses were prolonged, and erroneous keypresses were executed more slowly. However, the asynchronies between the hands did not increase in erroneous trials. 280 ms after erroneous keypresses a frontocentral positive potential was observed. In the following we will first discuss processes occurring before errors are committed and then turn to the processes occurring after errors are committed.
We assume that the ERPs elicited by incorrect performances reflect neural mechanisms that detect errors before they are actually committed, and before auditory feedback is available. Given the speed of movement sequences in the present study (about 3 keypresses with each hand per second), we suggest that internal forward models predicting the sensory consequences of actions 
are the basis for detecting the errors even before they were fully executed: Monitoring of fast movements, whose control cannot wait for sensory feedback, has to rely mainly on predictive (feedforward) mechanisms that compare internal action goals with the predicted consequences of planned movements.
Studies investigating the activity of neurons in the primary motor cortex (M1) of non-human primates showed that the latency between the first activity in M1 and movement onset is variable and can range up to several hundred milliseconds 
, but the typical assumed latency is around 100 to 150 ms (e.g. 
). At the same time as the motor command is sent from M1 to the periphery, an efference copy (or corollary discharge) is created in brain structures also involved in the generation of the movement. The efference copy is, however, not used to generate the ongoing motor activity, but can be used to predict the outcome of the motor command 
(information of efference copies interact at several levels of the central nervous system, and often modulate sensory processing; for reviews, see 
). The predicted outcome can be compared to the intended outcome, and an error signal is generated whenever there is a mismatch between intended goal and predicted consequence. The error signal can, in turn, modulate the motor command 
Accordingly, we assume that the mismatch between the predicted consequence of a planned keypress and the associated internal action goal, as detected by a feedforward control mechanism, is reflected in the pre-error negativity. From the present data we cannot conclude during which part of the movement (planning, initiation, early stages of execution) this feedforward control mechanism exactly occurs. However, it is important to note that a detection mechanism seems to operate before the pianists receive auditory feedback of their errors, i.e. before pianists perceive the auditory results of their actions.
The modulation of the motor command by the error signal of the feedforward mechanism might have resulted in the prolonged IOIs before and the slower velocities of incorrect keypresses, probably reflecting an attempt to avoid the error. In contrast to what one might have assumed, IOIs were not only prolonged for the hand that pressed the incorrect key, but IOIs were also prolonged for the other hand that pressed simultaneously the correct key. This is presumably due to bimanual coupling: studies show that bimanual movements begin and end at similar times, even when they have different parameters (e.g. amplitudes) and movement times differ when the respective movements are performed in isolation by one hand 
. Our task required tight bimanual coupling of the hands in terms of the timing. Correspondingly, asynchronies between the hands did not differ when an error was present or not.
One could argue that the pre-error negativity might reflect an error during memory retrieval and, thus, an even earlier stage than motor control or error monitoring. It is assumed that serial-ordering errors (i.e. notes that are intended at another location in the sequence) reflect the current activation of this erroneous element in memory 
. However, because pianists in our study performed the same tones in parallel with both hands (one octave apart), errors reflecting false memory retrieval should occur in both hands, instead of only in one. Because we only analyzed errors committed by one hand, it is unlikely that the pre-error negativity reflects false retrieval from memory. Moreover, one could also argue that the ERP difference before the note onsets might be due to motor-related processes. Motor execution processes are, however, expected to elicit lateralized EEG potentials 
, which is not consistent with our data: The separate analysis of left-hand and right-hand errors did not reveal any lateralization effect. Therefore, it is unlikely that the ERP difference reflects simply motor-related processes, but rather processes operating at a higher cognitive level, associated with monitoring or control. Finally, one could reason that the increased negativity before incorrectly played notes reflects a process that actually results
in the production of an error. For instance, a recent study 
showed that lapses in preparatory attention networks can lead to production errors. In that study the amplitude of the Contingent Negative Variation (CNV), a brain potential indexing preparatory attention, was decreased before stimulus presentation when an erroneous response occurred. Therefore, if lapses in preparatory attention were responsible for the errors in our study, one would have expected a similar decrease in ERP amplitude. However, ERPs elicited before incorrect performances had larger (negative) amplitude values than those elicited before correct performances, rendering such an explanation unlikely. Further, we think that the observed ERP difference in our study occurred too late to reflect lapses in attention. Considering the delay of activity in M1 to movement onset (presumably around 100 to 150 ms), lapses of attention should be observable before that time (as it was reported in 
), i.e. several hundred milliseconds before the button press. Thus, the fact that an increased negativity (instead of a decreased negative amplitude) was observed, in combination with the observed timing of the effect (around 100 ms before movement completion) renders it improbable that lapses in preparatory attention can account for the present findings. A similar explanation for the present results might be a temporal disengagement of the action monitoring system. Two other studies 
found that trials preceding erroneous trials (in Eriksen flanker and Stroop tasks) showed an enhanced positivity (compared to trials preceding correct trials), thereby ‘foreshadowing’ errors in future trials. This effect (termed the Error-preceding Positivity, EPP) is thought to reflect “transient deficiencies in the functioning of the monitor system prior to actual execution of an error” 
. These deficits may be associated with failures to activate adaptive control processes, resulting in occasional future errors. Because we observed no enhanced positivity before production errors, it is unlikely that a disengagement of the action monitoring system is reflected in the observed ERP effect.
The expertise of our participants and the characteristics of our task might explain why we did not observe an ERN (a potential frequently observed following the commission of errors, see 
for reviews) or an EPP component: In contrast to most studies investigating error processing (mostly in simple speeded response tasks, including the aforementioned studies 
) our participants did not react to external stimuli according to pre-defined arbitrary rules. Instead, they had to select the appropriate motor commands according to internal goals that they formed on the basis of instructions and the musical knowledge stored in their long-term memories. In addition, the present experimental situation reflects a task for which musicians are highly trained, compared to the button press responses to stimuli presented in standard error processing paradigms. Consequently, the error could be detected earlier than in choice reaction tasks. Incorrect notes also violated the regularity of the sequences and thus represented auditory oddballs, which are known to elicit a mismatch negativity (MMN; for a review, see 
). However, no MMN was visible in the ERPs, perhaps because it was overlapped by the positive potential emerging in a similar latency range (see below). Note that the magnitude of the ERPs (around 3 µV) was rather small compared to the amplitude of ERPs elicited in standard error processing paradigms 
. This is probably due to the complexity of our task, involving a range of interacting cognitive processes (e.g., memory retrieval, motor planning, performance monitoring etc., see Introduction). In addition, the simultaneous processing of input from different sensory systems (auditory, tactile, somatosensory) might have influenced the magnitude of the ERPs.
The fronto-central positive potential (emerging around 200 ms and) peaking around 280 ms after incorrect keypresses strongly resembles the Error Positivity (Pe), a potential frequently observed following the ERN in studies of error processing (for reviews, see 
). Although the functional significance of the Pe has remained rather unclear, three hypotheses about the Pe have emerged: The affective-processing hypothesis 
suggests that the Pe reflects affective processing of the committed error or its consequences. According to the behavior-adaption hypothesis 
, the Pe reflects the adaptation of response strategy after an error has been perceived, involving remedial performance adjustments following errors. The error-awareness hypothesis 
proposes that the Pe reflects the conscious recognition of a committed error. There is only little evidence in favor of the first two hypotheses, whereas there are some empirical data supporting the error-awareness hypothesis (e.g. 
). Another way of addressing the question about the functional significance of this potential is to consider its similarities to the P300 component, which has led to the suggestion that the Pe could reflect a P3b associated with the motivational significance of an error (for a review on the P300, see 
). The Pe, however, can be decomposed into an early and a late component, very similar to the distinction between P3a (indexing the involuntarily attention switch to novel and deviant stimuli, e.g. 
) and P3b (taken to reflect memory updating operations after task-relevant stimuli, e.g. 
, but see also 
). However, there are no studies directly comparing the early Pe with the P3a and the late Pe with the P3b, and therefore it remains unclear whether the early Pe reflects similar processes as the P3a. Based on previous studies 
we suggest that the positive deflection observed in the present study most likely reflects an early Pe or a P3a. Whether this potential is related to later processing stages of tactile and/or auditory feedback of the error, or simply due to the processing of an oddball stimulus (leading to an involuntary reallocation of attention) remains to be clarified. One way to address this would be to investigate performance errors committed in the absence of auditory feedback: if these errors also elicit the positivity, this potential cannot reflect auditory novelty processing.
In conclusion, the method of investigating motor experts in a natural context, accompanied with on-line measures of electrical brain activity (like EEG), can help to answer crucial questions in the domain of motor control and action monitoring. The occurrence of a pre-error negativity indicates that an early error detection mechanism operates in pianists even before an erroneous movement is fully executed. Our data also show that the early detection of errors influences movement execution, resulting in pre-error slowing of both hands and in keypresses with reduced velocity of the erroneous hand only. We assume that the underlying process is the detection of a mismatch between a predicted sensory consequence of an action and the intended action goal. Thus, our results reveal neural mechanisms that are able to detect errors prior to the execution of erroneous movements.