Appraisal Inference from Synthetic Facial Expressions

Abstract

Facial expression research largely relies on forced-choice paradigms that ask observers to choose a label to describe the emotion expressed, assuming a categorical encoding and decoding process. In contrast, appraisal theories of emotion suggest that cognitive appraisal of a situation and the resulting action tendencies determine facial actions in a complex cumulative and sequential process. It is feasible to assume that, in consequence, the expression recognition process is driven by the inference of appraisal configurations that can then be interpreted as discrete emotions. To obtain first evidence with realistic but well-controlled stimuli, theory-guided systematic facial synthesis of action units in avatar faces was used, asking judges to rate 42 combinations of facial actions action units on 9 appraisal dimensions. The results support the view that emotion recognition from facial expression is largely mediated by appraisal-action tendency inferences rather than direct categorical judgment. Implications for affective computing are discussed.