Classification of Natural Language Messages Using A Coordination Game

The role of natural language communication in economic exchange has been the focus of substantial experimental analysis. Recently, scholars have taken the important step of investigating whether certain types of communication (e.g., promises) might affect decisions differently than other types of communication (e.g., empty talk). Doing this requires the classification of natural language messages. Unfortunately, no broadly accepted method is available for this purpose. We here describe a coordination game for objective classification of natural language messages. The game is similar in spirit to the ‘ESP’ game that has proven successful for the classification of tens of millions of internet images. We illustrate our game using data reported by Charness and Dufwenberg (2006). We demonstrate that our objective classification procedure allows one to infer more from the data than is possible with subjective approaches