Connectionist mechanisms for cognitive control

An understanding of cognitive control is crucial for understanding high-level cognition and delineating the functional role of prefrontal cortex in supporting complex cognitive operations. In this paper, we approach the problem of cognitive control by examining the control needs of SHRUTI, a neurally plausible and cognitively motivated model of inference and decision-making. It is shown that processing via spreading activation has a number of limitations with respect to inference and decision-making, and specific forms of controlled processing is required to overcome these limitations. We propose a set of primitive, neurally plausible control mechanisms, including monitoring, filtering, selection, maintenance, organization, and manipulation, describe connectionist implementations of these primitive mechanisms, and demonstrate the use of several of these primitives in a complex control process.