Assessing the Efficacy of Open-Source Solutions to Automated Facial Coding: A Methods-Comparison Study with EMG. Face2face: advancing the science of social interaction.
Viswanathan, N., Labendzki, P., Perapoch Amado, M., Ives, J., Greenwood, E., Northrop, T., ... & Wass, S. 2022. Face2face: advancing the science of social interaction
Facial expressions are central components of face-to-face interactions and non-verbal communication. Most studies have measured changes in facial configuration by manually hand-coding videos of participants’ faces. This is relatively easy to setup and has facilitated theoretical developments in many fields e.g. spontaneous mimicry (REF). However, hand-coding does not track graded changes in action magnitude: they typically report onsets and offsets alone (REF). Without the action gradient, facial dynamics are reduced to binary events that do not differ in topography or temporal quality.
The ideal alternative, Electromyography (EMG), requires wired sensors be placed on skin. This spawns the possibility of participant discomfort, introduces the requirement for resources and limits ecological validity. Automated facial coding may provide an optimal trade-off: non-intrusive instruments that assess magnitude. Cross-correlations will be used to assess the degree to which the opensource auto-coder package (Mémoire IRCGN [https://github.com/LafLaurine/imac2-memoire-ircgn]) can approximate EMG data (Corrugator supercilii) in naturalistic, face-to-face interactions between mother-infant dyads (N - 10 dyads; infant age range 4-6 months). The videos will also be hand-coded enabling for the findings to link to the published literature.