Generating plural NPs in discourse : evidence from the GNOME corpus

Abstract

Research on the automatic generation of referring expressions has extended algorithms for the generation of full definite NPs to deal with anaphoric references. However, there has been relatively little work on the specific problems raised by plural anaphora. In particular, since plurals involve a reference to a set rather than an individual, one of the interesting questions concerns the extent to which elements of the set introduced by a plurality are salient, as compared to the plurality itself. This paper offers a preliminary exploration of these questions through a study of a small, richly annotated corpus. We discuss the implications of some of the findings for automatic referring expression generation, and also identify some fruitful avenues for future work.