Relying on a corpus of experimentally elicited task oriented conversations, we test the predictions of three different models of dialog - individualist, interactive alignment and interpersonal synergy – in assessing the efficacy of the linguistic exchanges. Employing Recurrence Quantification Analysis we quantify lexical, prosodic and speech/pause patterns in the conversations and show that the synergy model provides the best fit by being the most predictive of joint task performance.