Agreement and DP-internal feature distribution

An implicit assumption in most minimalist work is that DP as a whole carries all the φ-features with which external heads agree. In this paper I argue that under this assumption and the assumption that only a node that is φ-complete can delete the φ-features of a node with which it agrees, Chomsky's (2000, 2001) model of feature valuation is incompatible with a large body of work on the DP-internal distribution of φ-features, according to which neither N nor D enter the derivation being φ-complete. I consider several possible solutions and argue that this problem can most easily be avoided by adopting a feature-sharing model of the operation Agree, as proposed by Frampton & Gutmann (2006) and Pesetsky & Torrego (2007). Finally, several implications for Chomsky's theory of abstract Case are also discussed.