Negative polarity items (NPIs) like the English any and ever are thought to be licensed only if they occur in a downward entailing environment (one which supports inferences from sets to subsets). These NPIs are also licensed by affective predicates like surprise, sorry, and regret. I argue that these predicates are not downward entailing, and propose an alternative account of the licensing of NPIs in affective contexts. On my account, such NPIs are licensed by virtue of strengthening the informational content of the utterance they are a part of. I argue that such an account better can explain certain facts concerning the acceptability of NPIs in so-called “roofing” constructions.