Global positive polarity items and obligatory exhaustivity

I argue for a distinction between two types of positive polarity items (PPIs) which has not been recognized so far. While for some PPIs, anti-licensing is a strictly local phenomenon, for other PPIs anti-licensing should be stated as a global condition. I aim to contribute to a principled explanation for the distribution of a significant subset of global PPIs, by relating it to specific semantic properties of the relevant items. More specifically, I argue that PPIs such as soit . . . soit . . . , quelques and almost trigger obligatory exhaustivity effects and scalar inferences, and that independently motivated constraints regarding the generation of such inferences can account for their distribution. The paper also briefly addresses the case of other global PPIs, e.g., at least, for which a similar account is not straightforwardly available.