Should robots say sorry differently to humans? A multidimensional study of apology across task, severity, emotion, and nonverbal behavior in indonesia

Apologies play a critical role in restoring partners' trust in both human–human interaction and human–robot interaction (HRI). Humans express their apologies through diverse styles, emotional intensities, and nonverbal behaviors by adjusting to task contents and error severity. Robot may also be expected to have such abilities; however, the detailed requirement is still unclear. In this study, we conducted a multidimensional crowdsourced experiment in Indonesia to examine the social appropriateness of apology style, emotional intensity, and nonverbal behavior across multiple interaction contexts and error severities. The results show that apology styles need not be altered across error severities within the same task. Although not always the most preferred option, apologies that explicitly state a repair action were significantly selected across tasks. In contrast, emotional intensity and nonverbal behavior varied systematically with error severity across all tasks: neutral expressions and bowing were preferred for minor errors, whereas higher emotional intensity and culturally grounded Indonesian gestures were favored for major errors. These findings highlight that preferences for emotional intensity and nonverbal behaviors vary with error severity and provide an empirical foundation for designing adaptive apology behaviors in HRI.