A central finding in dialogue research is that interlocutors rapidly converge on referring expressions which become progressively contracted and abstract. However, there is currently no consensus on which mechanisms underpin convergence: The interactive alignment model (Pickering and Garrod) favours priming, the grounding model (Clark, 1996) prioritizes positive feedback, while Healey (2002) demonstrates the importance of miscommunication. To investigate convergence in closer detail we report a variant of the “maze-task". Participants communicate via a text-based chat tool which selectively transforms participants' private turn-revisions into public self-repairs that are made visible to the other participant. Dyads who received these artificially transformed turns used more abstract referring expressions, but performed worse at the task. We argue this is due to self-repairs having a beneficial effect of amplifying naturally occurring miscommunication (Healey et al., 2013), while also having a deleterious effect of decreasing participants' confidence in the conventions established during the task.