Antonio, Kate, and Harmony were all supposed to be in Ottowa right now for DH2020, but the pandemic shifted things online. Here is the lighting talk that Antonio was set to present on behalf of Dunham's Data. Kate and Harmony also recorded a talk about doing dance in a digital humanities context, for which the link is available via the DH2020 website.
The goal of this lightning talk is to present our approach, based on network analysis and visualization, to identifying and representing what constitutes movement communities in the project Dunham’s Data: Katherine Dunham and Digital Methods for Dance Historical Inquiry. The project uses the case study of Dunham’s company, in which almost two hundred dancers, drummers, and singers performed across the world for thirty years, and explores the role that movement sub-communities play in dance’s circulation and transmission within such a hyperconnected community of artists.
Computer-based solutions have previously been applied to model cultural and artistic transmission in the fields of theater and dance studies (Bench and Elswit 2016; Bollen and Holledge 2011; Caplan 2016). The challenge that the particular case of Dunham’s company poses is how to discover sub-communities among a hyperconnected community of artists who all worked together over time. In the mid-20th century, influence and transmission between performers occurs when they are in physical contact; consequently, we analyzed when performers coincide in the same place over time.
Through the concept of the “check-in,” that we borrowed from social media as a proxy for performance, we tracked what performers were present or absent in a specific place and day (read “Checking In”: The Flows of Dunham’s Performers for more information about check-in construction). In this regard, I built a customized sankey diagram that shows this circulation of performers in and out of the company. In Figure 1, company members are associated with a unique link (the colors will be explained below) that passes through or skips check-ins, represented by thin vertical bars. Figure 2 is an alternative visualization that explicitly plots check-in presences and absences in a timeline fashion.
Performers around shared check-ins depict a flow where the company’s embodied knowledge is potentially transmitted from individual to individual (watch the video Personnel Flow, 1947-1960 for further details). Based on this interconnectivity, I created a community network that displays the underlying social network shaped by the company’s interrelations. To do so, I worked with the same dataset used for the sankey diagram, removed the check-ins and connected directly every pair of performers who shared, at least, a check-in. Figure 3a shows the layout of this network at its first stage. Then, I applied a community detection algorithm to find potential sub-groups within the company.
Our key finding is the discovery of four* spatio-temporal communities, in which performers form a cohesive group with more connections to each other than to members of other groups. In Figure 3b, nodes’ color and proximity indicate the community they belong to (their size is proportional to their number of present check-ins). However, this categorical classification is not suitable for those nodes located in the center and between boundaries, as they seem to belong to more than one community at the same time. So I finally applied a community detection approach based on fuzzy clustering (Levner et al. 2007), that is, nodes do not belong to one single categorical group, but to more than one to a certain extent. Hence, border nodes between clusters may belong to more than one group but to a lesser degree (Pillar and Duarte 2010) than inner nodes (see Figure 3c).
Hyperconnectivity happens inside communities, while border nodes make visible inter-community connections across time. Central nodes are key performers in Dunham’s company and thus key to transmission across movement communities. They performed more times, with more performers, and for longer than the rest; they belong to multiple communities and they correspond to the nodes with the longest links in Figure 1, that is, they have passed through more check-ins than others. Figure 4 is an in-progress chord diagram that charts our community network with a circular layout. Notice that links in Figure 1 and nodes in Figures 3 and 4 use the same color scheme.
The flow diagram and the community network allows us to visualize the hyperconnectivity phenomenon from two complementary perspectives, in an effort to understand the complexity of this highly interconnected company. On one hand, the flow diagram shows lines of possible knowledge transmission between company members; and on the other hand, the community network shows how these members are structured in four different generations with some key performers as the “glue” holding the company together as a whole.
* fn: To be more specific, we should say that this is a non-deterministic algorithm that most times found four communities, but in some occasions found only three. This imprecision happens because the line separating two of them is too thin, that they both can perfectly be interpreted as a single community.