What number marking on indefinites means: conceivability presuppositions and sensitivity to probabilities

The inferences due to number marking on indefinites have been argued be pragmatic in nature, and to disappear in downward-monotonic environments. This paper shows that indefinites are associated to a conceivability presupposition even when embedded under negation. Furthermore, it presents the results of an experiment that shows, on the basis of a production task, that the preferred number for negated indefinites is sensitive to probabilistic information in a gradient way: the more common it is in general for the objects of interest to come in groups, the more plural is used, etc. The sketch of an account of these facts is provided, in a model of pragmatics where statements compete on the basis of their potential continuations.