Interpreting Human and Avatar Facial Expressions

Abstract

This paper investigates the impact of contradictory emotional content on people's ability to identify the emotion expressed on avatar faces as compared to human faces. Participants saw emotional faces (human or avatar) coupled with emotional texts. The face and text could either display the same or different emotions. Participants were asked to identify the emotion on the face and in the text. While they correctly identified the emotion on human faces more often than on avatar faces, this difference was mostly due to the neutral avatar face. People were no better at identifying a facial expression when emotional information coming from two sources was the same than when it was different, regardless of whether the facial expression was displayed on a human face or on an avatar face. Finally, people were more sensitive to context when trying to identify the emotion in the accompanying text.

Show Me or Tell Me: Designing Avatars for Feedback

“…They found that neutral expressions were more easily recognized on a human face than on an avatar. They also found that for the avatar faces, sadness was confused with the neutral face, supporting similar findings by Noël et al (2009).…”

“…They conclude that facial expressions cannot be considered in isolation. More recently, Noël et al (2009) described a study comparing emotions depicted in avatar facial expressions and short captions presented in speech bubbles. Five emotions were considered (anger, happiness, neutral, sadness and surprise) and congruent and incongruent presentations evaluated.…”

“…Depicting a neutral emotion can cause difficulties in interpretation. For example, Noël et al (2009) reported challenges in validating the cues they studied. The neutral avatar 4 Michelle Scott et al…”

A decade of NML networked graphics

“…a sad emotion for an angry statement) were still judged as having a less appropriate emotional expression but were not judged any less trustworthy, less convincing, or less sincere than the congruent case. There were no significant differences between the judgment of avatar faces and photorealistic faces [Noël et al 2009]. …”

Modelling human emotions for tactical decision‐making games

“…Carroll and Russell (1996) asserted that contextual information is crucial to identify or infer emotional states. To illustrate, a different emotion is attached to a crying avatar in Second Life when the accompanying chat message explains that the individual whom the avatar represents has either lost his job or got promoted (Noël, Dumoulin & Lindgaard, 2009). It thus seems plausible that emotion recognition is sensitive to the meaning of the scene in which a virtual human character appears.…”