Meaning and credibility in experimental cheap-talk games

We design experimental games to evaluate the predictive power of the first cheap-talk refinement, neologism-proofness. In our first set of treatments designed to evaluate the refinement with its usual emphasis on literal meanings, we find that a fully revealing equilibrium that is neologism-proof is played more often; senders deviate from an equilibrium in a way that can be predicted by the credibility of the neologism; and receivers' behavior indicates that they understand senders' deviating incentives. Our second set of treatments evaluates neologism-proofness from an evolutionary perspective in the absence of a common language. We find that the proportion of observations in which the meaning of a neologism evolves to disrupt a prevailing fully revealing equilibrium is lower when the equilibrium is neologism-proof. Our findings shed light on the capabilities and limitations of the refinement concept in predicting laboratory behavior under different language environments.