A central issue in the study of sentence processing is the manner in which various sources of information are used in resolving structural ambiguities. According to one proposal, the garden path model (e.g. Frazier & Rayner, 1982), perceivers are initially guided by strategies based solely on the structural properties of sentences. Another class of models, constraint satisfaction models, emphasise the influence of lexical properties in decisions among the alternative analyses of an ambiguous sentence fragment (e.g. Tanenhaus, Garnsey, & Boland, 1991). In this paper, we explore the prediction of an alternative model, the referential theory (e.g. Crain & Steedman, 1985).