Deriving person-based prominence effects: A set-based theory of AGREE

Person-based prominence effects occur when certain person categories or features are privileged by the grammar. This includes the PCC, split-ergativity, direct-inverse marking, omnivorous agreement, portmanteau agreement, and more. Since Silverstein (1976), this family of effects has been described by appealing to a Person-Animacy Hierarchy (PAH), which provides a ranking of person categories such as first, second, and third. The past 25 years have seen the advent of feature-centered approaches, such as the feature geometry of Harley & Ritter (2002). These theories stipulate entailment relationships between features such as [$\pi$], [participant], and [author] by necessitating that representations with more specific features such as [author] include less specific features such as [participant] and [$\pi$]. With a theory of agree that is sensitive to the geometry, the feature-based representation can be shown to capture the relationships between categories described by the PAH.

The first goal of this talk is to argue that both the PAH and the feature geometry are second order, in the sense that they describe properties that hold between categories or features, but should not be thought as being encoded in the representation itself. Besides having an undesirable representational status, the feature geometry in particular underpredicts the possible empirical range of prominence-based effects, as exemplified in this talk by the PCC and direct-inverse marking systems.

The second goal is to propose a new level of representation, termed primitives (as inspired by Harbour’s 2016 theory of person). These primitives combine into sets, which are linked to different features. These sets stand in subset-superset relationships, which allow the entailments between features to be captured through the first principles of set theory. This frees person from all second-order representations of entailment. I further show that reformulating agree to operate over these sets of primitives, rather than features or categories, allows all and only the possible range of person-based prominence effects to be captured, providing an empirical advantage over previous accounts.