[Part III. Semantics, pragmatics and discourse, Bare minimizers]
It is observed that Hebrew bare minimizers such as <i>kelev</i> (dog) differ from lexicalized minimizers such as <i>pruta leforta</i> (a red cent) in the following way. The former give rise to minimizing readings in negative environments but not in Yes/No questions, whereas the latter support such readings in both these environments. Thus, bare minimizers seem to have the same behavior as certain npis in other languages, which are licensed in some typical npi environments but not in others. The suggested explanation is this: While lexicalized minimizers have a covert even at LF, bare minimizers either do not have a covert even at all, or do not have a covert even with the right syntactic properties to induce wide scope alternatives.