Beyond DICE: measuring the quality of a referring expression
Kees van Deemter, Albert Gatt · OAR@UM (University of Malta) · 2009
This paper discusses ways in which the similarity between the contents of two referring expressions can be measured. Similarity metrics of this kind are essential when expressions generated by an algoritm are compared against the ones produced by human speakers, for example as part of an experiment in which referring expressions are elicitated. We discuss arguments for and against different metrics, taking our departure from the well-known Dice metric.