Investigating fragment usage with a gamified utterance selection task

Nonsentential utterances, or fragments, like A coffee, please! can often be used to communicate a propositional meaning otherwise encoded by a complete sentence I'd like to order a coffee, please!). Previous research focused mostly on the syntax and licensing of fragments, but the questions of why speakers use fragments and how listeners interpret them are still underexplored. I propose a simple game-theoretic account of fragment usage, which predicts that (i) listeners assign fragments the most likely interpretation in context and (ii) that speakers are aware of this and trade-off production cost and the risk of being misunderstood when choosing their utterance. Using a corpus of production data, empirically founded and precise model predictions are generated. These predictions are evaluated with two experiments using a novel gamified utterance selection paradigm. The experiments suggest that, as predicted, speakers take into account both potential gain in efficiency and the risk of being misunderstood when choosing their utterance.