Abstract
Facial expressions are behavioural cues that represent an affective state. Because of this, they are an unobtrusive alternative to affective self-report. The perceptual identification of facial expressions can be performed automatically with technological assistance. Once the facial expressions have been identified, the interpretation is usually left to a field expert. However, facial expressions do not always represent the felt affect; they can also be a communication tool. Therefore, facial expression measurements are prone to the same biases as self-report. Hence, the automatic measurement of human affect should also make inferences on the nature of the facial expressions instead of describing facial movements only. We present two experiments designed to assess whether such automated inferential judgment could be advantageous. In particular, we investigated the differences between posed and spontaneous smiles. The aim of the first experiment was to elicit both types of expressions. In contrast to other studies, the temporal dynamics of the elicited posed expression were not constrained by the eliciting instruction. Electromyography (EMG) was used to automatically discriminate between them. Spontaneous smiles were found to differ from posed smiles in magnitude, onset time, and onset and offset speed independently of the producer's ethnicity. Agreement between the expression type and EMG-based automatic detection reached 94% accuracy. Finally, measurements of the agreement between human video coders showed that although agreement on perceptual labels is fairly good, the agreement worsens with inferential labels. A second experiment confirmed that a layperson's accuracy as regards distinguishing posed from spontaneous smiles is poor. Therefore, the automatic identification of inferential labels would be beneficial in terms of affective assessments and further research on this topic.