Members of the Armed Forces are routinely faced with situations in which social interaction (e.g., cross-cultural, cross-linguistic) has both strategic and tactical implications. In these situations, anecdotal evidence indicates that some military members have greater outcome success (e.g., fewer subsequent fire-fights or IED events) when interacting with local citizenry than others, regardless of their individual language capabilities. In these increasingly common social contexts, where language barriers are often present, other modes of communication rise in importance. Nonverbal communication modes (e.g., body movement, physical proximity, gestures) and paralinguistic speech features (e.g., volume, pitch, turn-taking behavior), provide cues that carry significant meaning which can enhance cooperative interaction and build trust and rapport. This suggests that social interaction inherently involves latent communication features that are not easily discernible by an observer, and are similarly difficult for researchers to measure. However, these subtle "honest signals" (Pentland, 2010) are now measureable through the use of wearable sensors that capture body movement, proximity data, and speech features such as volume and pitch. The purpose of this paper is to introduce and test a conceptual model that captures the multimodal components of social interactions. An empirically validated model of social interactions will provide critical social skills content that will enable both training and evaluation. Data was collected using wearable sensors and observer assessments in a military training program designed to teach soldier leaders to be adaptive in unfamiliar environments. Training scenarios included interactions among soldiers and ethnic role players (who spoke languages other than English) acting as local citizens. Multimodal data were used to identify key features of social interaction that correlate with outcome measures. For example, body movement measures from the wearable sensors were found to be correlated with observer ratings of engagement.