It is more difficult to comprehend relative clauses formed on non-subject arguments than on subject arguments. However, this difficulty can be diminished if the non-subject argument is inanimate. In this paper we explore the hypothesis that relativized animates generate a specific, incremental expectation for a subject gap (Gennari & MacDonald, 2008). In two self-paced reading experiments using the filled-gap paradigm, we find support for the hypothesis that animate, but not inanimate, arguments are predictively linked to the subject gap. We explore the source of this expectation in a series of corpus and written production experiments. We argue it cannot be that the subject gap expectation follows from probabilistic knowledge conditioned on words or phrases per se, but that it must be conditioned on animacy as a grammatical feature.