Rational use of prosody predicts projection in manner adverb utterances

Speakers can be taken to be committed to utterance content even when that content is contributed in the scope of an entailment-canceling operator, like negation (e.g., Chierchia & McConnell-Ginet, 1990). We develop a probabilistic model of this phenomenon, called ‘projection’, that relies on the prosodic realization of utterances. We synthesize existing theoretical claims about prosody, information structure and projection into a model that assumes a rational speaker (Frank & Goodman, 2012) who produces utterances with prosodic melodies that can signal which utterance content she is committed to. Predictions of the probabilistic model are compared to the responses of an experiment designed to test the effect of prosody on projection in manner adverb utterances. Key behaviors of the model are borne out empirically, and the quantitative fit is surprisingly good given that the model has only one free parameter. Our findings lend support to analyses of projection that are sensitive to the information structure of utterances (e.g., Simons, Beaver, Roberts, & Tonhauser, 2017).