Multimodality is a virtually ubiquitous feature of communication. With the increasing interest in how animals, including humans, use multimodal and multicomponent signals in social interactions, there is an acute need for standardized and rigorous tools that will allow us to visualize, and analyze these signals as they occur in naturalistic interactions as a complex, integrated system. Network theory is a powerful methodology for intuitively visualizing and investigating the relationships between entities. Here, we propose a new application of network theory for analyzing multimodal communication. Using a case study of natural multimodal interactions in wild plains zebras (Equus quagga), we introduce the descriptive power of network metrics by providing an objective set of metrics to: (a) describe the relationships between simultaneously produced signals within and between modalities; and (b) infer signal meaning and function. Here we make available the multimodal communication data collected during our 2019 field season.

Ethical Note. This was a purely observational study conducted on free-living wild animals in Ol Pejeta Nature Conservancy (0°00 N, 36°56 E). Though the zebras in this population are habituated to the presence of humans in vehicles due to year-round tourist activity, we made efforts to minimize disturbance of natural behaviors during follows by ensuring the research vehicle never separated or obscured direct visual contact between harem-members, and by never redirecting natural harem movements (i.e., driving in front of the line of travel to induce a direction change). All research was conducted with the permission of Ol Pejeta Nature Conservancy and the Kenyan Ministry of Wildlife, and was in compliance with the guidelines of the Princeton University Institutional Animal Care and Use Committee (IACUC protocol #1835F).

Study Species and Subjects. We collected our data on free-living wild animals in Ol Pejeta Nature Conservancy (0°00 N, 36°56 E). We have been monitoring this population since 2001, using periodic census loops at an interval of 1 week (range: 1 day – 1 month) to collect data on associations, movement, and demography of individuals in the population. We are able to individually identify zebras based on their unique stripes. Data analysis was conducted on a focal subset of the population; 5 harems and one bachelor group. The harems consisted of 22 individuals, and group size ranged from 3 to 7 individuals (2 to 5 adult members). The focal bachelor group, KRT, comprised five individuals who were consistently in association across the study period.

Observations and Data Collection. Data collection occurred for 15 days between July 9th thru August 9th 2019 on the Eastern Sector of Ol Pejeta Conservancy. Observations occurred between 09:00-15:00, during which time a single focal group would be the subject of a focal follow for 2-4 consecutive hours. Interactions were captured with video using a Canon Vixia HR R700, and audio was recorded using a Sennheiser ME66 directional microphone in a Rode Blimp windshield and a Tascam Dr-100MKII Linear PCM recorder sampling at 48kHz. We adopted an all-occurrences sampling procedure in which we ad libitum captured interactions of individuals of the focal harem.

We amassed 45 contact hours during this study period. Of the collected footage, we only coded interactions in which the entire body of the focal individual was visible and unobscured, and eliminated videos in which the individuals were not in focus, partially out of frame, obscured by foliage or non-focal animals, or too distant for reliable coding of all regions of interest. This filtering yielded 13 hours of usable footage in which at least one interaction takes place involving a focal animal. We cleaned our sample data prior to analysis by removing all postures where one of the minimally necessary occurring visual units (i.e. ears, eyes, lower face, neck, tail) was not able to be coded. We also removed rare multimodal postures by eliminating those containing a component that was observed in < 2 interactions to improve the interpretability of our results by only considering components for which we had adequate co-occurrence data. After these cleaning steps, we were left with 1,594 high-quality multimodal postures.

Coding Scheme. For every interaction, we designated one individual as the focal individual and coded every signal produced by that animal for the duration of the interaction. If possible, dyadic interactions were coded twice, one with each interactant as the focal animal to capture both sides of an interaction, especially if the interaction was asymmetric (e.g., a dominance interaction). An interaction was defined as beginning when two individuals entered one body length of one another and the sender was 1) oriented towards the receiver, 2) the signal was produced in the direction of the receiver, 3) the sender approached the receiver, and/or 4) the sender made physical contact with the receiver. An interaction was defined as terminated when either 1) one individual moved beyond one body length of the other without being followed, or 2) the parties ceased interacting for at least 10 seconds (e.g., began to graze or sleep in proximity).