ABSTRACT
This chapter presents a novel connectionist architecture for modeling first impression formation, the process whereby a person makes judgments about another person based on limited observational data with no prior knowledge. It suggests how, by focusing on improving the fidelity of connectionist networks in reproducing human judgments, computational modelers can refine and make more accurate a psychological theory of how impression judgments are made. The specific psychological features, and corresponding additions to the computational model are prior probabilities, asymmetric connections, evaluation feedback and limited attention. The chapter presents a few key observations based on an analysis of simulations using present model and comparison models through human judgments, to illustrate some of the ways in which empirical data can be used to assess the significance of the architectural choices. An empirical finding that evidences the operation of a valence-driven process is one that suggests impression consistency ratings are far more sensitive to valence than they are to semantic relations.
