Abstract

While object coreference structures are possible in English, the equivalent German construction is marginal, and additionally seems not to obey Binding Condition B. Work on this thorny problem has been hindered by a Jack of agreement about which structures are and are not grammatical, and by the unpredicted nature of some contrasts. We applied the methodology of magnitude estimation (Bard et al 1996) to produce replicable grammaticality data as a basis for further work and to test competing accounts. The data reveals that none of the previous accounts we consider is wholly adequate. We show that the phenomena can be economically accounted for by a range of violable grammatical constraints on surface structure such as the Law of Increasing Members together with a violable form of Binding Condition B. We argue that the most adequate way of allowing these constraints to interact is in a constraint weighting model (Keller 2000), and show how such a model can be tightly constrained by deriving the constraint weightings directly from the experimentally obtained judgements.