Neural Mechanisms of Cognitive Control: An Integrative Model of Stroop Task Performance and fMRI Data Seth A. Herd, Marie T. Banich, and Randall C. O’Reilly Abstract & We address the connection between conceptual knowledge and cognitive control using a neural network model. This model extends a widely held theory of cognitive control [Cohen, J. D., Dunbar, K., & McClelland, J. L. On the control of automatic processes: A parallel distributed processing model of the Stroop effect. Psychological Review, 97, 332–361, 1990] so that it can explain new empirical findings. Leveraging other computational modeling work, we hypothesize that representations used for task control are recruited from preexisting representations for categories, such as the concept of color relevant to the Stroop task we model here. This hypothesis allows the model to account for otherwise puzzling fMRI results, such as increased activity in brain regions processing to-be-ignored information. In addition, biologically motivated changes in the model’s pattern of connectivity show how global competition can arise when inhibition is strictly local, as it seems to be in the cortex. We also discuss the potential for this theory to unify models of task control with other forms of attention. & INTRODUCTION Flexible cognitive control over our behavior is a key part of human intelligence. In what we call here the top- down excitatory biasing (TEB) model of cognitive control (e.g., Miller & Cohen, 2001; Cohen, Dunbar, & McClelland, 1990), the prefrontal cortex (PFC) is viewed as maintaining representations that guide control of tasks. These PFC representations provide an excitatory top-down bias to groups of neurons processing task- relevant information. Because their activity is heightened relative to neurons processing task-irrelevant informa- tion, distracting information has less effect (Corbetta, Miezin, Dobmeyer, Shulman, & Petersen, 1991). This explanation is consistent with Desimone and Duncan’s (1995) biased competition model of attention—TEB theory explains task control as another form of atten- tional control. This theory has the virtues of simplicity and accords with a great deal of data, but it does not address the nature and origin of these task representations in the PFC. For example, in the neural network Stroop task model of Cohen et al. (1990), it is simply assumed that the PFC has existing representations tuned for the task of naming ink colors. These color-naming task rep- resentations provide extra input to the color-naming processing areas in the posterior cortex so that they out-compete the stronger (more practiced) word-reading pathway. This extra input (top-down excitatory bias) is what supports the ability to identify the ink color of a word even when that word names a different color (e.g., ‘‘red’’ printed in green) (the Stroop task). Previous TEB models cannot account for patterns of brain activation observed in fMRI studies of the Stroop task. Of note, more activation has been observed in brain regions responsible for processing word-related in- formation on incongruent (‘‘red’’ printed in green) than neutral trials (‘‘lot’’ printed in green) (Banich, Milham, Jacobson, et al., 2001; Banich, Milham, Atchley, Cohen, Webb, Wszalek, Kramer, Liang, Wright, et al., 2000b). Previous TEB models predict that activity in word- related brain regions should be less than that in re- gions processing color information, and that this effect should happen equally in incongruent and neutral trials. To resolve this inconsistency between theory and data, we posit that category representations are involved in cognitive control. When first called upon to identify the ink color, already existing category information about ‘‘color’’ is used to guide attentional control. This information is likely not to be specific to ink color per se, but to apply to the general category of color (although with time and practice it may be honed more specifically to ink color). Interestingly, research with monkeys indicates that the same areas of the PFC that are involved in top-down control are also involved in cate- gory representations. For instance, neurons in the PFC of monkeys distinguish between category boundaries (e.g., dog vs. cat) and the response of these neurons University of Colorado Boulder D 2006 Massachusetts Institute of Technology Journal of Cognitive Neuroscience 18:1, pp. 22–32