Implications of grammatical gender for the theory of uninterpretable features

I argue that grammatical gender is semantically empty but intrinsically valued, so the strict linkage between uninterpretable and unvalued in Chomsky (2001) cannot be correct. I then demonstrate that gender is infinitely reusable as an “ activity” feature; in contrast, abstract Case activates a DP for just one Agree relation. This asymmetry suggests that valuation via Agree causes goal deactivation, and that deactivation is not necessary for every uninterpretable feature (uF). I accordingly analyze deactivation as arising from PF illegibility of multiple values for a single feature. Agree relations value Case, but never value nominal gender, so the legibility problem does not arise. I demonstrate that in Bantu, adjunction of N to D makes gender accessible to all probes outside DP. This and the reusability of gender as an activity feature leads to a cluster of systematic contrasts between Bantu and Indo-European languages: Bantu DPs A-move much more freely than Indo-European DPs, and value iterating subject agreement. The facts thus demonstrate that the internal syntax of DP impacts its feature matrix; it is not the case that a DP automatically inherits all f-features of its subparts, as syntactic theory generally assumes. Finally, I illustrate that Bantu C and T can agree with different expressions, casting doubt on the Feature Inheritance approach to uF in Chomsky (2007, 2008) and Richards (2007). The facts of grammatical gender argue that valued uF Transfer to the Conceptual-Intentional Interface without inducing crashes.