The availability of data on digital traces is growing to unprecedented sizes, but inferring knowledge from large-scale data is far from being trivial. The field of computational social science uses these large datasets to test theories from the social sciences, aiming at a quantitative understanding of various social phenomena. I will illustrate how the digital traces of online communication can be used to test theories from sociolinguistics and psychology in two examples. First, I will show an analysis of how human-computer interaction shapes languages, testing the relationship between the keyboard layout and the sentiment expressed and elicited in various digital media. Second, I will present an analysis of linguistic intergroup biases, measuring how concreteness and sentiment in Wikipedia biographies depends on demographic and cultural factors.