Motion-Capture Patterns of Voluntarily Mimicked Dynamic Facial Expressions in Children and Adolescents With and Without ASD

Emily Zane, Zhaojun Yang, Lucia Pozzan, Tanaya Guha, Shrikanth Narayanan, Ruth B. GrossmanJournal of Autism and Developmental Disorders 2018 · DOI

Introduction

In social interactions, humans rely on subtle, dynamic changes in facial features to interpret the emotions and intentions of others. For individuals with Autism Spectrum Disorder (ASD), these interactions are often fraught with difficulty. Previous research has consistently noted that neurotypical (NT) observers perceive the facial expressions of those with ASD as "awkward," "unnatural," or "ambiguous." However, identifying the precise physical characteristics that lead to these perceptions has remained a significant challenge.

Traditional methods for studying facial expressions, such as human coding or facial electromyography (fEMG), have limitations. Human coding is inherently subjective, while fEMG typically only measures electrical activity in a few specific muscle groups, failing to capture the full complexity of facial movement. This study, a collaboration between researchers at Emerson College, USC, and the University of Warwick, addresses these gaps by employing high-resolution 3D motion capture (MoCap) to objectively quantify the geometry and timing of facial mimicry in real-time.

Methodology

Data Acquisition and Processing

The researchers utilized a VICON MoCap system, capturing data at 100 frames per second, to track 32 reflective markers placed on the faces of participants. To move beyond raw coordinate data, the team calculated nine specific Euclidean distances—such as eye opening, mouth widening, and face lengthening—to represent various muscle contractions and facial configurations.

To account for variations in individual face shapes and sizes, the data underwent a normalization process where individual mean coordinates were aligned with a global mean. This ensured that the observed differences were due to movement patterns rather than anatomical differences.

Statistical Innovation: Growth Curve Analysis (GCA)

A key contribution of this paper is the application of Growth Curve Analysis (GCA). Unlike traditional analyses that focus solely on the "peak" intensity of an expression, GCA uses polynomial vectors (linear, quadratic, and cubic) to model the entire trajectory of a movement over several seconds. This approach allows researchers to capture the nuances of how an expression builds, is sustained, and eventually subsides.

Experimental Design

The study involved 19 children and adolescents with ASD and 18 neurotypical peers, matched for age, gender, and IQ. Participants were asked to voluntarily mimic 36 short, silent video stimuli from the "Mind Reading CD," which featured actors portraying a range of universal emotions. These stimuli were independently validated by 22 adults to establish ground truth for valence (positive vs. negative) and intensity (high vs. low).

Key Findings

Lack of Emotional Differentiation (Valence Blindness)

One of the most striking findings was the difference in how valence influenced facial movement. In neurotypical participants, the magnitude of facial movement was highly predictable based on whether the emotion was positive or negative (Estimate = 0.71, p = 0.002). In contrast, for participants with ASD, valence did not significantly predict movement (p > 0.10). They tended to move their faces to a similar degree regardless of the emotional content, suggesting a lack of physical differentiation between distinct emotional states during mimicry.

The Sustainability Gap

The analysis of movement trajectories revealed a significant "sustainability gap." While NT individuals were able to maintain high-intensity expressions for several seconds, participants with ASD exhibited much steeper negative linear slopes (Estimate = -0.80, p < 0.0001). This indicates that while they could reach the intended expression, they struggled to sustain its intensity, leading to a rapid decay of the facial pose.

Micro-bursts and "Jerky" Affect

The MoCap data provided an objective basis for the perceived "awkwardness" of ASD expressions. Quantitative analysis of acceleration profiles revealed that the ASD group’s movement is marked by narrow, frequent spikes in acceleration across time, whereas NT movement is smoother and more fluid. Furthermore, the ASD group exhibited significantly higher variance in facial distances (SD = 7.8) compared to the NT group (SD = 2.0). These brief, spasmodic bursts of movement—reminiscent of micro-expressions or motor tics—create a "jerky" quality that likely contributes to the ambiguity and atypicality perceived by social partners.

Discussion and Conclusion

This research demonstrates that the "awkwardness" often attributed to facial expressions in ASD is rooted in quantifiable, dynamic movement patterns. The findings of valence blindness, poor sustainability, and increased jerkiness provide a new framework for understanding social communication challenges in autism.

Limitations and Future Directions

While this study focused on volitional mimicry, the authors acknowledge that spontaneous, natural expressions may show even greater impairment and should be the focus of future MoCap research. Additionally, while the current analysis averaged distances across the entire face, future work will investigate whether specific regions (e.g., the eyes versus the mouth) are primary drivers of atypicality. A follow-up study is currently underway to directly link these objective MoCap patterns to the subjective judgments made by neurotypical observers, further bridging the gap between physical movement and social perception.

Summary updated

BibTeX
@article{zane2018motion,
  author = {Emily Zane and Zhaojun Yang and Lucia Pozzan and Tanaya Guha and Shrikanth Narayanan and Ruth B. Grossman},
  title = {{Motion-Capture Patterns of Voluntarily Mimicked Dynamic Facial Expressions in Children and Adolescents With and Without ASD}},
  journal = {Journal of Autism and Developmental Disorders},
  year = {2018},
  doi = {10.1007/s10803-018-3811-7},
}
Figure