Paper: Learning Preferences for Referring Expression Generation: Effects of Domain, Language and Algorithm

ACL ID W12-1503
Title Learning Preferences for Referring Expression Generation: Effects of Domain, Language and Algorithm
Venue International Conference on Natural Language Generation
Session Main Conference
Year 2012
Authors

One important subtask of Referring Expres- sion Generation (REG) algorithms is to se- lect the attributes in a definite description for a given object. In this paper, we study how much training data is required for algorithms to do this properly. We compare two REG al- gorithms in terms of their performance: the classic Incremental Algorithm and the more recent Graph algorithm. Both rely on a notion of preferred attributes that can be learned from human descriptions. In our experiments, pref- erences are learned from training sets that vary in size, in two domains and languages. The results show that depending on the algorithm and the complexity of the domain, training on a handful of descriptions can already lead to a performance that is not significantly different from training on a much...