Attention and the processing of context: A parallel distributed processing approach to normal and disordered cognition

This thesis addresses the cognitive mechanisms underlying attention and the processing of context, their disturbance in schizophrenia, and their relationship to biological variables and structures. It is made up of three papers, each of which focuses on a separate component of this research (all three were written in collaboration with authors whose names appear on the individual papers). In PAPER 1, "On the control of automatic process: A parallel distributed processing model of the Stroop effect," a model is presented that is used to account for a variety of the empirical phenomena associated with normal performance in a standard measure of selective attention. The model provides an explicit mechanism for selective attention, as well as a precise account of the relationship between selective attention and the processing of context. PAPER 2, "Context, cortex and dopamine: A connectionist approach to behavior and biology in schizophrenia," explores schizophrenic deficits in these areas. Three types of tasks are addressed: the Stroop task, the continuous performance test (a measure of signal detection), and lexical disambiguation. Three PDP models are described that simulate the performance of normal and schizophrenic subjects in each of these tasks. The same disturbance introduced into all three models is shown to induce schizophrenic patterns of performance in each of the tasks. This disturbance involved a change in the gain parameter of units in the module responsible for representing and maintaining context in each model. It is argued that this provides a model for the effects of dopamine in prefrontal cortex, relating disturbances of this system to schizophrenic patterns of cognitive deficit. PAPER 3, "A network model of neuromodulatory effects on signal detection: Gain, signal-to-noise ratio, and behavior," elaborates the argument that changes in gain can be used to model the effects of catecholamines such as dopamine. This paper undertakes a direct analysis of the effects of the gain parameter in simple networks, and compares these to data concerning the biological and behavioral effects of catecholamines. The overall goal of this work is to bring the theoretical methods of cognitive psychology to bear on understanding disturbances of cognition in schizophrenia; and to draw, in particular, upon the principles of parallel distributed processing to relate these disturbances to underlying biological mechanisms.