Moving together: Interpersonal Coordination and Individual Identification in Dyadic Dance

Abstract

We move our bodies as a response to listening to music. These responses could range from simple movements such as head-banging, finger-tapping, and feet-tapping to a complex dance. Dance often occurs in social settings, where individuals entrain their movements to the music being played along with the visual cues derived from others. This phenomenon is termed rhythmic social entrainment. Moving in synchrony with others has been shown to foster social bonding. Dyadic dancing emerges as a first step in investigating rhythmic-social entrainment. This thesis examines the dyadic dance context from two angles: interpersonal coordination and individual identification. We first predict perceived interaction and similarity using kinematic and gestural features. Then, we explore the existence of unique signatures of individuals within dyads amidst interpersonal coordination. As an extension, we also look at the presence of unique movement signatures in a markerless-choreographic setting. Interpersonal coordination has been studied using two perceptual variables: Interaction and Similarity. Studies have identified many postural and gestural features that exhibit moderate correlation with perceptual variables. However, several aspects of interpersonal coordination remain underexplored, particularly the role of musical features and energy levels of individuals in dyads. This thesis addresses these gaps by investigating the influence of music’s danceability, which is primarily characterized by pulse clarity on interpersonal coordination, revealing a very strong statistically significant correlation between danceability of musical stimulus and mean perceived interaction ratings across all dyads dancing on that stimulus. This finding highlights the facilitative role of danceable music in enabling coupling. Furthermore, this study explores the link between the energy levels of dancers and interpersonal coordination, demonstrating perceived similarity is associated with similar energy levels within a dyad . In addition to this, energetic dyads are also perceived as more interactive, likely due to the impression of enjoyment and engagement they convey and vice versa. As a final step, we take dyad recordings and attempt to classify the perceptual variables into three classes: low, medium, and high. We employ novel features, such as Energy and Covariance Matrix, in addition to the ones from the literature, to train the model. We achieved reasonable accuracies in predicting “Interaction” and “Similarity.” We also examined the joints that are important in the classification of these variables. This analysis revealed the significance of hands in predicting interaction relative to other body parts, which is consistent with other modalities, including spoken communication. Carlson et al. [17] demonstrated the existence of motoric fingerprints by identifying individuals dancing freely to the music stimulus using only the movement features with notably high accuracy. It is interesting to examine whether an individual has a unique signature even while dancing with a partner in the presence of interpersonal coordination. We achieved noteworthy dancer identification accuracy, signifying the existence of motoric fingerprints in the dyadic contexts. In addition to this, we demonstrated the joint pairs and joints that are key to the classification model. We employed the dyadic model to predict individual dancers based on features extracted from their solo performances. The high identification accuracy achieved indicates a strong consistency of unique movement signatures across both solo and dyadic settings. However, our misclassification analysis identified certain individuals who were not correctly predicted by the dyadic model. This anomaly was explained in terms of empathy dynamics of individuals within dyads. Studies in the domain of music-induced movements predominantly rely on marker-based methods for movement capture. However, these methods suffer many limitations, with a primary concern being the potential alteration of natural movements due to the presence of markers on subjects. This poses a threat to the ecological validity of such studies. We examined the markerless data of professional dancers executing the same choreography. We investigated the notion of the personal style of a dancer by training a dancer identification model based on movement features. We achieved a dancer identification accuracy at least two times higher than the chance level, signifying the existence of motoric fingerprints in a choreographic-markerless setting. In summary, this thesis contributes novel insights into interpersonal coordination in dyadic dance. It also shows the presence of motoric fingerprints in dyadic as well choreographic dance contexts, verifying the external validity of Carlson et al. [17]’s methods in dancer identification from movement features and explores the applicability of markerless movement data.