I introduce two new sets of data which appear to be problematic for the standard assumption that selection is restricted to syntactic elements in a sisterhood relation (a fact that follows as a theorem from certain versions of Merge). The first set comes from the well-known domain of Warner/Lasnik/Potsdam facts, aka "*I'm America, and so can you!". Lasnik famously argued that peculiarities of English verbal morphology were responsible for this pattern, and work by Potsdam argued that the pattern could be accounted for better by a constraint on head-movement out of antecedents of ellipsis. I show, based on novel data from English-Greek bilinguals, that neither approach is correct: the same patterns are found when the antecedent VPs are in Greek. But head movement does play a role in generating new parallelism relations, though these relations are not the same as typical variable-binding configurations. Second, I examine a new set of data from a database I'm building of cross-categorial l-selectional in English: while the majority pattern seen in e.g. rely_V on, reliance_N on, reliant_A on points to the conclusion that l-selection is a property of roots, independent of the category those roots surface in, a significant minority pattern such as pride_V oneself on, pride_N in, proud_A of surprisingly demonstrate categorially sensitive l-selectional properties. I argue that this kind of selection only appears to be nonlocal, but can in fact be analyzed as a result of the categorizing head selecting the internal argument; I further argue, on the basis of historical changes in cross-categorial mismatches in selection, that l-selection is stochastic.