Languages vary as to what kind of phrasal categories allow recursive iteration of self-same embedding. Children first learn an embedding rule, then must learn whether the rule can apply recursively or not. However, direct experience of recursive embedding is rare in the input. A study of recursive nominal modification in Spanish show children acquire different types of modification (possession, part-whole relations) at different times even if these are expressed with the same preposition de. This suggest that the domain of rule formulation is narrower than syntactic category (PPs) or even lexical particle (de). Bilingual children show delays in acquiring a first level of embedding rule but not in allowing the rule to be recursive. This suggests that learning recursive modification is not sensitive to the concomitant reductions in input in bilingual contexts. I argue that children learn that embedding rules are recursive by inference from the productivity of simple embedding rules. The evidence on the acquisition of recursive nominal modification points to the limitations of the parameter setting model of syntactic development.
