Gently does it: Humans outperform a software classifier in recognizing subtle, nonstereotypical facial expressions.

Area of Science:

• Psychology
• Affective Science
• Nonverbal Communication

Background:

• Dominant theories propose innate, universal facial expressions for emotions.
• Existing research relies on intense, stereotypical expressions, rarely found in real life.
• Subtle and non-prototypical emotional expressions are common but understudied.

Purpose of the Study:

• To develop and validate a novel stimulus set of dynamic, subtle emotional facial displays.
• To compare the recognition and naturalness of subtle versus prototypical facial expressions.
• To evaluate the performance of software classifiers and human experts on subtle expressions.

Main Methods:

• Experiment 1: Developed and validated a new set of subtle, dynamic facial expressions.
• Experiment 2: Tested subtle and prototypical expressions using a software classifier and human expert coders.
• Stimulus validation involved human ratings for recognizability, ecological validity, and naturalness.

Main Results:

• Subtle expressions were well recognized by humans and rated as more naturalistic and ecological than prototypical expressions.
• Software classifiers trained on prototypical expressions performed poorly on subtle expressions.
• Human expert coders noted fewer key facial movements in subtle expressions compared to prototypical ones.

Conclusions:

• Emotions can be effectively communicated through subtle, non-prototypical facial expressions.
• While prototypical expressions are recognized, they may not reflect real-world emotional communication.
• Current automated systems struggle to interpret the nuances of subtle emotional displays.