Using robustness to learn to order semantic properties in referring expression generation

Abstract

A sub-task of Natural Language Generation (NLG) is the generation of referring expressions (REG). REG algorithms aim to select attributes that unambiguously identify an entity with respect to a set of distractors. Previous work has defined a methodology to evaluate REG algorithms using real life examples with naturally occurring alterations in the properties of referring entities. It has been found that REG algorithms have key parameters tuned to exhibit a large degree of robustness. Using this insight, we present here experiments for learning the order of semantic properties used by a high performing REG algorithm. Presenting experiments on two types of entities (people and organizations) and using different versions of DBpedia (a freely available knowledge base containing information extracted from Wikipedia pages) we found that robustness of the tuned algorithm and its parameters do coincide but more work is needed to learn these parameters from data in a generalizable fashion.