Refer efficiently: Use less informative expressions for more predictable meanings

We present the results of a large-scale web experiment investigating comprehenders’ ability to guess upcoming referents in an unfolding discourse. Participants were given a text that had been cut off just before a noun phrase, and attempted to guess which previously mentioned referent, if any, would be mentioned next. Our results show that writers are more likely to refer using a pronoun or proper name rather than a full NP when comprehenders have less uncertainty about the upcoming referent, and are more likely to use names than pronouns when comprehenders all tend to makes guesses to one or a few incorrect referents. These effects hold beyond other possible influences on the choice of referring expression type. Our results support addressee-oriented accounts of referring form choice (e.g. Brennan & Clark, 1996; Arnold, 2008) and suggest that language is a rational solution to the problem of communication: shorter and less informative expressions are favoured when less information is sufficient to carry the message (e.g. Jaeger, 2006; van Son & Pols, 2003).