Defeasibility is a central concept in non-monotonic reasoning. In this paper we connect the peculiar looseness of non-strictly-deductive inference - looseness grounded on "default" or defeasible assumptions - to the looseness of referential uses of definite descriptions. In both cases we are faced with the problem of revising some premises of an argument and changing the conclusion. We aim to show that inferential defeasibility requires attention to lexical and cultural context dependence and to different forms of arguments, also depending on whether premises are given as general propositions or singular propositions. Although our main point is on different aspects of defeasible arguments, the paper is also an attempt to link problems typically discussed in the philosophy of science with problems discussed in the philosophy of language.