“In Study 3, we tested the accuracy of these forecasts by randomly assigning participants to speak for 30%, 40%, 50%, 60%, or 70% of the time in a dyadic conversation.“

This study does not correlate to the real word at all. A computer telling you when you should speak is so different from a natural conversation. In the real world, you’re not speaking because you’re listening, intentionally; in the study, you’re not speaking because the computer said who’s turn it was.