On Collection and Covert Variables

It has become commonplace in semantic theorizing to argue that the semantic representation of certain linguistic expressions contains covert elements in addition to what is contributed by the overt linguistic material. Theorists have pursued this strategy with respect to a wide range of constructions in a wide range of languages. A small sample of such expression types in English would include comparative adjectives (covert delineation of comparison), quantifiers (covert domain restriction), and event reports (covert location/time indexes). For this kind of move to be suitably constrained, semanticists have developed a number of tests that are supposed to give independent evidence about whether there are in fact the covert elements postulated. These include, inter alia, binding (Partee 1989; von Fintel 1994; Stanley and Szabó 2000; Stanley 2002), optionality (Recanati 2004), control (Bhatt and Pancheva 2006), sluicing (Merchant 2001), and collection (Cappelen and Lepore 2006; Cappelen and Hawthorne 2009). This note is about one of these tests in particular: the collection test. We don‘t want to deny the need for covert variables in the semantics that receive their values from context somehow; on the contrary, we assume (without argument) that there are such variables. Rather, our specific aim in this paper is to ask whether the collection test is a successful