The role of agreement in clitic-doubled constructions

Summary and ConclusionWe began with the premise that doubling with DO- and IO-CLs requires that both elements of the chain agree in features. With IOs, doubling is universal because the features of the IO argument are practically irrelevant (section 1.1). However, since DO-CLs are inherently specific, this feature is crucial for DO-doubling to be grammatical. Animacy is the second feature necessary in most contexts. However, as pointed out in section 1.3.1, doubling may operate even in the absence of the latter. The same matching of features required of doubling constructions is required under extraction (a subset of doubling structures), hence, no special mechanisms are needed to reject ill-formed sentences. What causes ungrammaticality in extractions from CL-D DOs is a clash between the specific referential CL and a nonspecific interpreted object argument, a violation of the Matching Principle. Since lexical partitives and unagreement phenomena are productive ways to signal the specificity of the DO, extractions at LF and in the syntax proper can produce well-formed sentences (section 2.2). Moreover, the Matching Principle follows directly from the unique indexing peculiar to chain coindexing and Spec-head agreement.These results have two main consequences: (a) the hypothesis that CLs are agreement morphemes which do not absorb Case becomes viable. This claim is supported by the doubling of inanimates (section 1.3.1) and even of some animates (17) in the absence of a, by facts about Case with IOs (section 1.3.2), and by weak cross-over effects (section 3.1) where CLs-as-agreement serve to identify the relevant empty category. (b) Since extractions of both CL-D IOs (section 2.1) and CL-D DOs (section 2.2) are possible, it follows that CLs are not theta-role absorbers, because the doubled constituent must be in an argument position. The moral is that the absence of extraction does not itself show that the double is in an A' position. Rather one must look deeper to discover whether independent principles in the language can account for lack of extraction in such contexts.For extensions of our main hypothesis to other Spanish dialects see Suñer (1986). Finally, the assumption that datives in a CL-chain are NPs and not PPs (see Appendix) makes possible the generalization that contemporary Spanish CLs — whether direct or indirect — enter only into nominal chains, a generalization confirmed by their parallel behavior with respect to weak crossover and scope (sections 3.1–3.2). This conclusion brings the analysis of Spanish in line with those of French and Rumanian (Steriade 1980/81) where datives have been argued to be NPs, and it opens the door for the possibility that datives in all the Romance languages might be nominal in nature.Interestingly, datives are also NPs in Hebrew (Borer and Grodzinsky 1986).