Evaluating Centering for Information Ordering Using Corpora

• DOI: 10.1162/coli.07-036-r2-06-22
• Publication Date: 2008-10-15T14:16:11Z
• Authors: Nikiforos Karamanis; Chris Mellish; Massimo Poesio; Jon Oberlander
• Keywords: Text generation; Baseline (sea)

In this article we discuss several metrics of coherence defined using centering theory and investigate the usefulness such for information ordering in automatic text generation. We estimate empirically which is most promising metric how useful a general methodology applied on corpora. Our main result that simplest (which relies exclusively NOCB transitions) sets robust baseline cannot be outperformed by other make use additional centering-based features. This can used development both text-to-text concept-to-text generation systems.