Generating Easy References: the Case of Document Deixis

This paper argues that algorithms for the generation of referring expressions should aim to make it easy for hearers and readers to find the referent of the expressions that are generated. To illustrate this claim, an algorithm is described for the generation of expressions that refer across a hierarchically ordered domain, and which takes search effort into account by adding logically redundant information. To support the ideas underlying the algorithm, a psycholinguistic experiment is presented that confirms readers’ preference for the generated, logically redundant expressions over non-redundant alternatives.