Referring Expression Generation (REG) from sets of attributes addresses the following question:
[G]iven a symbol corresponding to an intended referent, how do we work out the semantic content of a referring expression that uniquely identifies the entity in question? (Bohnet & Dale, 2005)
The input to a REG from AS system typically consists of a several sets of attributes (e.g. {type=lamp, colour=blue, size=small}), where at least one attribute set is labelled the intended referent, and the remainder are the distractors. In the traditional formulation, the system's task is to output a set of attributes for the intended referent that uniquely distinguishes it from the distractors, while being as small as possible.
The traditional task formulation has been extended in various ways, including realisation of attribute sets as NPs and use of corpus-based evaluation methods.
Legacy resources from the TUNA Shared Tasks (2007-2009):