Evidence from neuroimaging and lesion studies indicates that both medial temporal lobe (MTL) and prefrontal cortex (PFC) contribute to episodic memory performance. We present a computational neural network model of how interactions between MTL and PFC support performance on free recall tests. The MTL component has already been used to account for a wide range of recognition memory findings (Norman & O'Reilly, in press). The PFC component is based on recently developed models of how PFC supports cognitive control. These models (e.g., Frank, Loughry & O'Reilly, 2001) posit that PFC actively maintains aspects of presented stimuli via multiple parallel 'stripes' that be updated separately. On each trial, information in PFC can be maintained, or can be replaced by aspects of the current stimulus. Over time, the pattern of activity in PFC can be viewed as an evolving 'context vector'. At study, the current state of this PFC context vector is associated with item representations via the hippocampus, and at test PFC can cue memory for studied items. At a high level, this approach has much in common with more abstract models of temporal context memory (e.g. Howard & Kahana, 2002a). We use the computational model to generate specific testable predictions about the performance of normal and frontally damaged subjects in a variety of free recall paradigms. Modifications to the model are proposed that will allow us to capture a wider range of behavioral data, including primacy and subjective organization. This framework allows us to develop a mechanistic understanding of prefrontal contributions to episodic memory. ? ?
