Connectionist Models of Emotional Distress and Attentional Bias

Abstract

It is proposed that connectionist models may contribute to the theory of attentional bias associated with emotional distress. The principal features of the bias shown in studies of the emotional Stroop test that simulations should capture are reviewed. Three broad mechanisms for simulating bias are identified: (1) high exposure to emotional stimuli during learning; (2) supernormally high intensity of emotional input; and (3) activation of task demand units associated with response to emotional stimuli. The suitability of these mechanisms for modelling emotional Stroop test findings was explored using a modification of the architecture developed by Cohen, Dunbar, and McClelland (1990) to investigate the standard Stroop test. Manipulation of activation of a “threat-monitoring” task demand unit provided the pattern of performance closest to existing data from distressed individuals: an emotional Stroop effect on colour-naming of moderate magnitude, a large standard Stroop effect, and a minor facilitative bias in reading emotional words. Training of the threat-monitoring unit on either all emotion words, or on a single emotion word only, allows modelling of the interference effects associated with both general emotional distress and phobias. Generalisation of the effect across different emotional stimuli varied with the generality of the threat-monitoring task. Advantages and limitations of this connectionist approach are discussed.