Capturing spontaneous interactivity: a multi-measure approach to analyzing the dynamics of interpersonal coordination in dance improvisation

Abstract

Introduction: Interpersonal coordination is widely acknowledged as critical to relating with, connecting to, and understanding others, but the underlying mechanisms of this phenomenon are poorly understood. Dance-particularly improvised dance-offers a valuable paradigm for investigating the dynamics of interpersonal coordination due to its inherent ability to connect us. However, conventional approaches to studying coordination often fail to capture the co-creative spontaneity that is intrinsic to such interactions.

Methods: This study combined multiple measures of interpersonal coordination to detect moments of high coordination between two freely improvising dancers. We applied maximum correlation vectors, normalized Symbolic Transfer Entropy (NSTE), and surveys to analyze the time-varying dynamics of similarity in movement speeds, directed influence, and subjective perception of dancers engaged in an improvisation task.

Results: This multi-measure approach offers a means of capturing the interplay between different dimensions of interpersonal coordination.

Discussion: This approach may be used to understand the underlying mechanisms of co-creative social interactions in improvised dance and other forms of spontaneous interactivity.

Keywords: coordination dynamics; interactional synchrony; movement; pose estimation; social interaction; symbolic transfer entropy; video analysis.

Figure 1

Experimental setup. The participants, labeled here with their respective referents (A,B) are seated perpendicular to one another at a square table. Markers (yellow and black stickers) are placed on their hands for post-experiment pose estimation using DeepLabCut (open-source pose estimation software). Improvisation task structure is included in bottom left.

Figure 2

Illustration of the segmentation and commentary form structure and the data types collected from each segment.

Figure 3

Demonstration of DeepLabCut pose estimation of user-defined body parts. Note that multiple body parts were labeled (as seen in this figure) to train a robust neural network, but only the hand data was used for analysis.

Figure 4

Overview of each coordination measure derived from a single session (dyad 1). Dotted lines delineate the boundaries of each perceptual condition. Moments of rated high coordination are denoted by red bars, while moments of rated low coordination are denoted by blue bars. To enhance signal clarity, the NSTE and asymmetry plots have been smoothed using non-overlapping 3-s windows.

Figure 5

(A) Four representative moments illustrating true positive and true negative instances. For each moment, the maximum correlation vector is depicted alongside a vignette of the corresponding segment of interaction. High coordination moments are highlighted by orange panels and low coordination moments by blue panels. Solid grey lines denote the unique significance threshold for each moment while the dotted lines represent the classifier threshold y = 0.5. (B) Confusion matrix summarizing the results of classification.

Figure 6

MCV, NSTE, and NSTE asymmetry from representative moments of high coordination (orange) and low coordination (blue). MCV significance threshold is indicated by the dotted line, while the classifier threshold is indicated by the solid line.

Figure 7

Cross-section of MCV, NSTE, and NSTE asymmetry from four moments of high coordination sufficiently overlapping with participants’ segments, alongside commentary provided by participants.