A Yawning Journey: A Study on Contagious Emotions via Facial Electromyography

Why we smile? As many simple questions, the answer may be all but simple. As a first assumption, we could answer: 'Sometimes we smile because we see the smile of others'. Then why we yawn? Apparently, not only for similar reasons, but also using muscles in a similar way. The same logic applies to pain, but in the upper part of our face. Mirrored smiles, yawns and painful expressions will be the main topic of this analysis, because the ability to recognize an emotion exclusively on the visual information of somebody else's face is an essential feature for communication and social interaction in a society structured not only on materials and wealth. In order to comprehend the complexity of emotional mimicry, it is crucial to acquire information on two main aspects of a person: the psychology of the subject and the biological response, acquired trough electromyography (EMG - principal focus of this paper) and electrodermal activity (EDA). It is important to underline that these three contagious expressions were part of a bigger and more ambitious project revolving also around three social emotions and the construction of a database of videos of spontaneous complex emotions. Of these social emotions, many express problems already in translation attempts and, hopefully, will be exhaustively detailed in future projects: - embarrassment (that has a proper translation in many languages and needs no further introduction); - 'disprezzo' (the proper translation of this Italian word is not immediate. 'contempt' or 'disdain' are close, but 'disprezzo' also implies a feeling of superiority); - 'shadenfreude' (which has no direct translation neither in English nor in Italian and expresses a form of joy for the misfortune of others). A total of more than 60 people, mostly university students, volunteered to be part of this project and the process of data acquisition (in a shared neuroscience laboratory) lasted for more than two months. The project, still continuing and evolving today, was divided in 3 main parts (referred as 'phase 0', 'phase 1' and 'phase 2') and only the third part involved the use of EMG's techniques. The EMG's signals were acquired with non-invasive surface electrodes, strategically placed to capture the electrical activity of three key facial muscles: the zygomaticus major, widely activated during smiles, the corrugator supercilii, usually associated with expressions of pain, sadness and anger, and the orbicularis oculi, which plays a crucial role in differentiating various types of smile, for instance the joyful one and the 'shadenfreudeful' one. The signals were preprocessed, then used to identify the highlights of the video footage in order to visually detect and manually label the subjects' expressions in an faster way (compared to the complete vision of the video footage). The signals were then subjected to data augmentation algorithms, feature extraction algorithms, principal components analysis (PCA) and classified using various methods of machine learning: support vector machine, random forest, decision tree and k-nearest neighbors. Results show a classification accuracy, in some cases, higher than 95%, but the complete extent of the results must also consider the variability of the subject-based train/test set division and computational costs.