Dual-Level Connectionist Models of Group Cognition and Social Influence

Elements and Processes

Regardless of the specific applications, DLCMs contain a common set of components: nodes, links, dynamic processes, and modules. Nodes are perceived communication events (inputs), mental states associated with them (processing nodes such as beliefs, policies, and emotions), and communication outputs connected to mental states (arguments, emotional statements, etc.). For example, in a DLCM theory of argumentation and influence in groups, input nodes might be policy-relevant arguments from other group members as perceived by a particular member, the set of policy options relevant to those (processing nodes), and a set of arguments produced by that group member and transmitted to the other group members, and so on for all group members. These nodes are connected to each other, although not fully, and are the potential sites of change, both communicative and mental. Thus, DLCMs are connectionist because of the connections between the nodes and dual-level since change can occur in both the communicative and cognitive levels.

Connections between nodes, or links, have constraints, valences, and activation weights. Constraints determine the connections between types of nodes. For instance, in the example above, links run (a) from perceived messages (inputs) to policy options, (b) among policy options (processing nodes), (c) from policy options to generated arguments, and (d) from members' arguments to other members' received messages. In a DLCM group argument theory, (a), (c), and (d) are unidirectional flowing forward as described above. Unidirectional links, for example, can run from perceived messages to mental states but not the reverse. Thus, a theoretical constraint is placed on types of links.

Valences specify the negative or positive implications of the links between nodes. For example, a negatively valenced link between a mental state, say an emotion, and an emotional message indicates that the more an emotion is experienced, the less likely it is to be expressed—feeling [Page 323]anger leads to the repression of expressed anger. A positively valenced link indicates the reverse. Moreover, each link has an activation weight that indexes the strength of a link. Stronger arguments for a policy option have larger activation weights for links to policy options than do weaker arguments.

...