Agreement and Concord in Nominal Expressions

Feature sharing is a pervasive property of natural languages that has long been considered quite puzzling. (cf. Barlow & Ferguson 1988:1-4). Being the source of redundancy, it is obviously a debated issue in the minimalist approach which aims to reduce language to a conceptually necessary system (cf. Chomsky 2005). In this paper, I deal with two different notions of feature sharing as they are manifested inside nominal expressions (NE): Agreement (the relation between an argument and the head noun, with the effect of (genitive) Case assignment) and Concord (the relation between adjectival modifiers and the head noun). Despite both are mediated by functional projections containing uninterpretable features, I claim that these two relations are crucially different. Agree arises from the merger of a formal uninterpretable feature which is needed in the extended projection of the head and is checked against a lower projection which may move or not. Concord arises from the merger of a modifier which is itself underspecified for uninterpretable features and never gives rise to movement. The empirical domain is provided by the two unrelated families of Romance and Bantu languages, as the rest of the contributions of this volume. As for all natural languages, micro-variation can be found in strictly related languages, while macro parallelisms can be drawn between two languages belonging to unrelated families. This latter approach is not so often entertained as the former. My contribution is mostly focused on Romance but will observe the Romance language under consideration (Romanian) under the novel perspective of the comparison with Bantu. On the other hand, it will also propose a novel account of pre-prefixes in Xhosa and a general account of double definiteness observed both in Bantu and in Romanian. As for all the work in the generative enterprise, the ambition is that the general principles discussed here hold for all natural languages subject to different parameters which should derive all and only possible variation (in the space limits of a paper). In section 1, I introduce to the general framework. In section 2, I argue that Agree in the NE also targets person features as it does in the clause. I also observe that in Romance and Bantu these features are weak. As a consequence the probed constituent overtly moves to the Spec of the Agreeing head only if it does not contain any lexical element. In section 3, I distinguish Agreement from Concord showing that they can co-occur. In section 4, I further specify the notion of Concord and discuss possible applications to pre-prefix spreading and double definiteness. In section 5, I discuss open questions and possible consequences of the analysis.