Localization of a Neural System for Error Detection and Compensation

Gehring, Goss, Coles, Meyer, and Donchin (1993) have reported electrophysiological evidence for a brain mechanism dedicated to monitoring performance and compensating for errors (see also Falkenstein, Hohnsbein, Hoormann, & Blanke, 1990). They have described a component of the human event-related potential, called the error-related negativity (ERN), which is characterized by a negative peak about 100 ms following the onset of electromyographic (EMG) activity when the subject is in error on that trial. The amplitude of the ERN is larger when the subject strives for response accuracy than when the subject concentrates on speed. It is also correlated with several error-compensation variables. The larger the ERN, the smaller the force with which the erroneous key is pressed, the higher the probability of correcting the error by immediately depressing the other key, and the slower the reaction time on the next trial. These correlations suggest that the neural system whose activity is reflected by the ERN is involved in the active inhibition and correction of an error as soon as it is detected. Gehring et aI. (1993) noted that the data from their study did not allow for the localization of this neural system in the brain. They recorded from five electrode sites and observed only that the ERN was largest over the front and middle of the scalp. They did, however, offer some speculation, based on animal evidence, pointing toward "a system involving the anterior cingulate cortex and supplementary motor areas" (p. 389). Our own data allow for a direct confirmation of this localization in humans. We recorded high-density event-related potentials using a 64-channel geodesic electrode net (Tucker, 1993), with an interelectrode spacing of about 4 cm. This methodology enabled us to fully characterize the scalp topography of the ERN in two different experiments. In the first experiment (Dehaene, 1994), 12 subjects were presented on each trial with a single Arabic or spelled-out numeral. They had to press one key with one hand if the target was larger than 5 and another key with the other hand if it was smaller than 5. In the second experiment (Dehaene, unpublished data), 12 other subjects were presented with lists of words, which they had to classify as belonging or not to a target semantic category (e.g., an-