Talking Together, Working Apart: Predicting Collaborative Problem Solving Performance Using Language

Collaborative problem solving (CPS) skills are critical for students and workers in the 21st century. However, reports show that students and workers routinely underperform at CPS-related tasks. While many studies have investigated the factors that contribute to CPS performance, few have focused on prediction, and even fewer have focused exclusively on language. This study takes a unique prediction-first approach, where the goal is to identify the features of language that best predict CPS performance, and then use those linguistic features to build explanatory models of CPS performance. Overall, we found that more sophisticated content words indicate worse CPS performance, while more sophisticated function words indicate better CPS performance. Additionally, we saw that teams using more concrete content words performed worse at the CPS task, while teams using more abstract content words performed better. Finally, we found that teams performed better when using positive emotion words (especially positive nouns) and words indicating high arousal.