Projection without lexically-specified presupposition: A model for 'know'

Abstract

We present an analysis—formalized as a computational cognitive model—of the projection of the content of the clausal complement of know in utterances of negated declaratives. Our model, formulated within the Bayesian Rational Speech Act framework (Frank and Goodman, 2012), derives projection from lexical entailments of know and sensitivity to the Question Under Discussion (QUD; as do Abrusán, 2011, and Simons et al., 2017), as well as reasoning about utterance informativity relative to private speaker assumptions (Qing et al., 2016; Warstadt, 2022). Crucially, our model predicts projection for know without encoding the inference via a lexically-specified constraint on the Common Ground. The model goes beyond existing analyses by also making predictions about the projection of the content of the clausal complement of nonfactive think, as well as other types of inferences. We find support for three qualitative predictions of our model in two experiments that measured the projection inferences participants draw about utterance content.