@InProceedings{ Pacheco_al:12,
author = {Pacheco, Fabian and Duboue, Pablo and Dominguez, Martin},
title = {On The Feasibility of Open Domain Referring Expression Generation Using Large Scale Folksonomies},
booktitle = {Proceedings of the 2012 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies},
month = {June},
year = {2012},
address = {Montr\'{e}al, Canada},
publisher = {Association for Computational Linguistics},
pages = {641--645},
url = {http://www.aclweb.org/anthology/N12-1082}
}
Quality referring expressions help distinguish a referent from other, similar entities in the discourse. Mixing text from different sources can lead to failed referring expression, as in the example above (it should be the US president). There are many well understood algorithms for referring expression generation but they need ontological data. In this paper, joint work with Grupo PLN FaMAF, we explored the use of an ontology derived from Wikipedia to do referring expression generation. We found the ontology to have information about the entities frequently but the traditional algorithms from the field to produce poor referring expressions given that data.
This paper started a number of papers on this data and problem.
On a personal note, the paper was the undergraduate diploma thesis for a student that came back to the university after a long hiatus. He took my NLG course and from there we end up with this communication.
The paper is available at the ACL anthology. The slides are also available.
A later talk at gave at Interactions NJ contains a more refined discussion on the topic.