Effective communication in project teams is important, but not often taught. We explore how feedback might improve teamwork in a controlled experiment where groups interact through chat rooms. Collaborators who receive high feedback ratings use different language than poor collaborators (e.g. more words, fewer assents, and less affect-laden language). Further, feedback affects language use. This suggests that a system could use linguistic analysis to automatically provide and visualize feedback to teach teamwork. To this end, we present GroupMeter, a system that applies principles discovered in the experiment to provide feedback both from peers and from automated linguistic analysis.