Learning to Communicate in Cheap-Talk Games

We analyze simple adaptive learning processes to model the evolution of effective communication in cheap-talk games with initially meaningless signals. We suggest that learning rules may be sensitive to the information available to players at the population level. As a consequence, learning outcomes may depend on this information. We show that more information can be a barrier to comunication if new messages are introduced by mistake. If on the other hand the source of new messages is deliberate experimentation, then more information is not detrimental to communication.