Language is an imperfect and coarse means of communicating information about a complex and nuanced world. We report on an experiment designed to capture this feature of communication. The messages available to the sender imperfectly describe the state of the world; however, the sender can improve communication, at a cost, by increasing the complexity or elaborateness of the message. Here the sender learns the state of the world, then sends a message to the receiver. The receiver observes the message and (...) provides a best guess about the state. The incentives of the players are aligned in the sense that both sender and receiver are paid an amount which is increasing in the accuracy of the receiver’s guess. We find that the size of the language endogenously emerges as a function of the costs of communication. Specifically, we find that higher communication costs are associated with a smaller language. Although the equilibrium predictions do not perform well, this divergence occurs in a manner which is consistent with the experimental communication literature: overcommunication. We find that the sender’s payoffs, relative to equilibrium payoffs, are decreasing in the cost of communication. We also find that the receiver’s payoffs, relative to equilibrium payoffs, are increasing in the cost of communication. Finally, we find imperfections in coordination on the basis of the experimental labels. (shrink)