Strategic regulation of cognitive control by emotional salience: A neural network model

Abstract

We present a neural network model of stimulus processing, which uses a mechanism of adaptive attentional control to regulate the moment to moment deployment of attention according to both the demands of the current task, and the demands of emotionally salient information. This mechanism allows negative emotional information to reduce cognitive control to aid in the detection of threats, which produces a momentary withdrawal from the current task set to allow unbiased processing of available information. The combination of cognitive and emotional regulation of task set allows this model to address inter-trial aspects of emotional interference in colour naming. In particular, we focus on the nature of the emotional interference in colour naming (McKenna & Sharma, 2004) as well as in word reading (Algom, Chajut, & Lev, 2004) and show how this form of interference is functionally distinct from the classic Stroop effect. Our model addresses a range of findings in colour naming and word reading tasks and is informed by recent neuroimaging data concerning the interaction between the anterior cingulate and prefrontal cortices. The model is used to explore the interface between cognition and emotion with a series of predictions, including a qualitative distinction between state and trait forms of anxiety.

Acknowledgements

This work was funded by EPSRC grant GR/S15075/01 and a grant from the Gatsby Charitable Foundation.

We thank Kiran Kalindindi, Su Li, Phil Barnard, Eddy Davelaar and Daniel Algom for insightful discussion and comments.

Notes

¹Error trials in our model would also be associated with elevations in the conflict measure.

²The increase in the slow component for state anxiety during word reading is reduced because the word-processing bias nears the asymptotic maximum of the logistic function in this implementation.

³This same model has also been implemented with the more biologically plausible equations described by O'Reilly (2001). Both versions of the model are available in Matlab code by e-mail request from the authors or online at: http://www.bradwyble.com/research/models/cogcontrol/.

⁴These are given as default values for those layers not mentioned below.

⁵Negative bias pushes nodes to negative values by default.

⁶This elevated threshold value for inhibition within the category layer ensured that there was minimal inhibitory competition even when category nodes were excited by bias input.