There are at least four opposing views on the directionality and configuration of Agree relations. In mainstream Minimalism, Agree is strictly downward “probing” (Chomsky 2000, 2001, Boskovic 2007, Epstein & Seely 2006), but some recent works argue instead that Agree always looks upwards (Zeijlstra 2012; Wurmbrand 2012). A third perspective takes agreement to follow from the Spec, head configuration (Koopman 2006; and see Chomsky’s 2013 labeling through shared prominent features), and under a fourth proposal, the directionality of Agree varies parametrically (Baker 2008). While each of these approaches has some empirical support, none achieves the strict conceptual necessity that is a central goal of Minimalist theory. In this paper I argue, following Epstein’s 1999 approach to c-command, that a downward directionality for Agree follows automatically and necessarily from the bottom-to-top construction of syntactic objects. But if a uF does not find any source of valuation in its c-command domain at first Merge, directionality-free matching and DELAYED VALUATION are possible up to the point of cyclic Transfer. The approach eliminates several stipulations from agreement theory, deducing that the uFs making goals “active” in Agree relations are simply those that find no match in their c-command domains at Merge.
Carstens, Vicki. "Delayed valuation: a reanalysis of goal features, “upwards” complementizer agreement, and the mechanics of Case." Syntax 19, No. 1 (Jan 2016): 1-42. doi:10.1111/synt.12116.