Yichen Wu¹ (), Jonas Nölle¹, Chaona Chen¹, Oliver Garrod¹, Robin Ince¹, Philippe Schyns¹, Rachael Jack¹; ¹University of Glasgow

Human society relies on maintaining healthy social bonds. When transgressions are made against others, expressing apology is critical to successfully repairing those bonds. Yet, current understanding about apology signals remains controversial. Some theories suggest that apology signals should comprise negative emotions, such as sadness or shame, while others argue that smiles are necessary for appeasement. To resolve this debate, we modelled dynamic facial expressions of apology using a 3D dynamic generative model of the human face and the data-driven method of reverse correlation. On each experimental trial, we generated a facial animation comprising a random selection of facial movements (i.e., Action Units, AUs), eye gaze direction (up, down, left, right, direct) and blushing (present/absent) displayed on a randomly generated face identity (white, 18–35 years, male/female). Participants (N = 60, white Western European, Mage = 20.6 years, sex-balanced) rated how sorry each person looked in response to a given transgression (e.g. “He stepped on your foot”) on a 5-point scale from ‘Not sorry at all’ to ‘Extremely sorry’. Each participant completed 2400 such trials in a between-subjects design (n = 20 per transgression) with sex of stimulus face blocked and counterbalanced. We then modelled the specific facial cues of apology by quantifying the statistical relationship between each facial cue (e.g. presence of an AU) and the participant’s responses using Mutual Information and non-parametric permutation testing to establish the statistical threshold. Results showed that facial expressions of apology comprise frowning, lateral lip stretching, smiling, and averted gaze (e.g., looking down); blushing did not have a significant effect. Together, our results show that communicating apology involves a unique constellation of facial movements that reflect a combination of negative emotions and appeasement, thereby unifying current accounts. Future work will examine whether and how apology signals vary across transgression types, relationship types, and cultures.