Even the most cursory observation of the world’s languages reveals a fairly robust correlation between structural cases and agreement. Thus, it is very often the case that subject agreement targets noun phrases that surface as nominative, and object agreement targets noun phrases that surface as accusative. And note: the existence of non-nominative subjects and non-accusative objects demonstrates that this is not merely a terminological truism. Let us refer to this interdependence of case and agreement as the Case-Agreement Codependence Generalization.
Most of the explanations offered for this generalization have followed one of two general paths. First, it has been suggested that structural cases (like nominative and accusative) arise as the result of the noun phrase having undergone agreement (George & Kornfilt 1981; Chomsky 2000, 2001). Second, it has been suggested that there is some independent factor – e.g. occupying [Spec,IP] – that conditions both structural case and agreement, giving rise to the observed correlation (see, e.g., Chomsky 1986). But there is a third logical possibility: that structural case on a noun phrase is a precondition for that noun phrase being targeted for agreement. This third possibility has recently been argued for, e.g., by Bobaljik (2008).
In this talk, I discuss a few systematic exceptions to the Case-Agreement Codependence Generalization, and argue they favor a version of this third approach – i.e., that structural case is a precondition for agreement – where cases like “nominative” and “absolutive” are nothing but the absence of otherwise valued case on the noun phrase.
On this view, it is the absence of valued case that makes, e.g., a “nominative” noun phrase a viable agreement target. While this bears some similarity to Chomsky’s (2001) Activity Condition, it has several distinct advantages with respect to the data we will explore. First, it accounts for the occurrence of structural case in the absence of agreement. Second, it allows noun phrases bearing structural case to enter into more than one agreement relation. Third, it accounts straightforwardly for instances where targets of finite agreement go on to be assigned other case markings, later in the derivation.