Ellipsis really is binding
Simon Charlow, NYU

Abstract
There's some powerful arguments that ellipsis requires some sort of interpretive identity between antecedent VPs ('α') and the ellipses they license ('ε') (e.g. Sag 1976). However, in the past 25 years, the field has converged on the view that simple interpretive identity cannot be what relates α and ε (e.g. Evans 1988); specifically, the interpretive identity approach under-generates. Coping with this state of affairs has resulted in a complicated, inherently non-compositional hybrid characterization of ellipsis licensing which ensures that α and ε have (a) some limited form of syntactic identity, while at the same time (b) giving rise to a coherent discourse pair (Rooth 1993). But against this hybrid view, ellipsis sites behave essentially like structureless pro-forms whose reference is supplied via the same general mechanisms which undergird anaphora resolution (Hardt 1999; Schwarz 2000; Charlow 2008; 2012). In a slogan: it seems that in some cases α binds ε. Indeed, I'll argue that new data warrant a stronger conclusion: all ellipsis is binding of ε by α; thus ellipsis resolution is a totally semantic process. Yet ellipsis as binding seems difficult to square with the arguments that enforcing interpretive identity between α and ε under-generates. Though this seems like a paradoxical state of affairs, I offer an account on which it isn't. We can have a purely semantic theory of ellipsis as binding which covers all the data in a principled way.