Abstract

Abstract In this paper, we caution that the comparative is, in fact, not, a licensing environment for NPIs. We show that the appearance of NPIs is much more restricted than previously assumed: strong NPIs do not appear in comparatives, and often NPI- any is confused with free choice any . Strong NPIs are licensed only if an antiveridical function is introduced, such as the negative metalinguistic comparative charari (Giannakidou and Yoon 2009)—but the comparative itself does not contain an antiveridical or downward entailing operator. Importantly, NPI sanctioning in comparatives is limited to rescuing (Giannakidou 2006) which allows only the weakest NPI type, the one that can be sanctioned in violation of LF licensing. The implication of our analysis is that the comparative should not be thought as a licenser of NPIs—a fact consistent with the analytical difficulty, admitted in many works, in making the comparative downward entailing or nonveridical. Finally, it cannot be claimed that the comparative contains negation—if it did, strong NPIs should be fine, but they are not