Regularities in null instantiation

Analyses of null instantiation phenomena in English have tended to look for a unifying explanation covering all cases, for instance, in terms of aspectual structure (Rappaport-Hovav & Levin 1998; Wright and Levin 2000) or selectional restrictions (Resnik 1993). However, as argued by Goldberg (2005) and also pointed out earlier by Fillmore (1986), null instantiation does not seem to be a uniform phenomenon. There are lexical idiosyncrasies–predicates with similar meanings nonetheless differ in whether they allow omission of a given argument (eat versus devour); differences between instances of null instantiation in how the unexpressed referent is interpreted, anaphorically (I know 0 ) or existentially (I was eating 0 ); and differential effects of context on omissibility (This lion has killed 0 before versus #He nearly killed 0 ). In this paper I show that drawing careful distinctions between omission types allows us to state regularities that capture the insights and intuitions behind the single factor analyses while respecting idiosyncrasy where necessary. The two main distictions to be made in the characterization of omissions are (1) the type of licensor–lexical versus constructional–and (2) the interpretation of the omitted referent–anaphoric versus existential. Two major regularities involving these criteria can be observed. First, among lexically licensed omissions, the interpretation of an omitted argument correlates with the frame semantics (Fillmore 1985) of the predicate. Second, the two interpretation types, whether licensed lexically or constructionally, are compatible in distinct ways with a speaker’s need to produce an informative utterance. With anaphoric omissions a specific referent–a token– is recoverable and the core participants of the event or relation are fully specified. In the case of existential omissions, the omitted referent is de-emphasized for one of two reasons, which are not necessarily mutually exclusive. The focus of interest is on the action itself, as illustrated by data discussed in Goldberg 2005, or what is relevant is the state that one of the instantiated participants, typically the subject, is in as a result of being, or having been, involved in the specified event type. The remainder of this paper is structured as follows. Section (2) presents the classification of omission types in terms of licensors and interpretation. Section (3) makes the case for the alignment of framal semantics and interpretation type. Section (4) discusses the broader interaction of communicative intent and the use of argument omission with a given