Probe-Goal (Chomsky 2000, 2001) purports to account for clausal domain agreement as well as structural case, but does not address case and agreement uniformity within the DP. In Russian, the determiner, adjective, and noun all share gender, number, and case features. Previous analyses of agreement within the DP do not account for case-sharing (Carstens (2000, 2001)) or they rely on additional mechanisms restricted to the DP (Wechsler&Zlatić (2000), Franks (1995)). I propose to address these problems by modifying the Agree relation as follows: a Probe Agrees with all active Goals in its domain until the end of the phase. This modification offers better empirical coverage than Carstens (2000, 2001), while avoiding the split clausal/nominal approach to agreement (Wechsler&Zlatić (2000), Franks (1995)).
Under the assumption that Adj and N enter the derivation with interpretable [+I] agreement/φ-features and uninterpretable [-I] Case feature, consider two potential instantiations (1-2) of Probe-Goal, both of which are problematic. In (1) [TP T0(-I) [DP D0(+I) [AP A0(+I) [NP N0(+I)]]]], D has [+I] φ-features. T and D Agree, but T is barred from additional Agree relations because it becomes inactive after the first instance of Agree and can no longer act as a Probe for N and Adj. Neither can D serve later as a Probe for N or Adj, because D would have become inactive as well. Additionally, its [+I] φ-features prevent it from being a Probe. In (2) [TP T0(-I) [DP D0(-I) [AP A0(+I) [NP N0(+I)]]]], D has [-I] φ-features and probes/Agrees with the Adj or N, but later it cannot act as Goal of T. In both instances, the initial probe would be inactive after the first Agree operation.
I argue for a modified form of Agree in which a Probe may continue valuing case until the end of the phase (and in which (1) is the derivation that is most viable under Probe-Goal). Given my hypothesis that T enters into Agree with each case-marked head to value abstract Case, both case and φ-feature uniformity are accounted for.
Given phases (Chomsky 2001), Agree is also constrained within a phase. In a transitive construction, I propose that v values the Accusative case of each head within the object DP similar to case assignment for the subject as outlined above. At phase end, the features that have been valued in that phase are Spelled Out and barred from future Agree operations. Further in the derivation, T can (and must) Agree with all matching and still active Goals. As the object DP has already been sent to PF, it is no longer an eligible Goal. Thus, the sentential subject is the only DP that can enter into Agree with T, having its nominative Case valued. This allows multiple Agree to be powerful enough to license Case and uniform features to any number of case-marked heads within a DP, but prevents it from applying to DPs that entered into Agree in an earlier phase.