In this talk I (re)consider two broad typological generalizations which indicate that agreement and case are fundamentally distinct types of operation, and that the dependency between them is asymmetrical, a conclusion that runs contrary to the widespread view that case and agreement mark 'two sides of the same coin' (Chomsky) or head vs. dependent marking of a single abstract relation (Nichols).
(i) Dixon (1994) and others have noted that case and agreement may mismatch in their alignment, but only in one way. Languages are well-attested with an ergative/absolutive case alignment but a nominative/accusative [=subject-object] agreement alignment, but the reverse is claimed not to occur. Bobaljik 2008 offers an account of this asymmetry (cf. Baker 2008) which crucially rejects the 'two-sides' view of case and agreement. The talk revisits that argument, arguing that the generalization survives despite subsequent challenges (Legate, Deal).
(ii) Case and agreement also differ in broad terms in the prevalence of active (Split S, Fluid S) alignments, far more robustly attested in agreement patterns than in case patterns. The Dependent Case Theory (DCT, Marantz 1991, Baker 2015), invoked in the account of (i), may predict this distribution. Under this perspective, agreement is a relationship between NPs and functional heads (thus an active pattern is readily described), but case represents a relationship among NPs in a given domain. On this view, the absolute position of a single NP should play no role, thus an active pattern requires additional machinery. We offer a somewhat tentative exploration of ways in which even the best-described active case patterns show evidence of further complexity beneath surface appearances, consistent with the expectations of DCT.
(Portions of this talk include joint work with Mark Baker.)