Paper: Statistical Generation: Three Methods Compared And Evaluated

ACL ID W05-1601
Title Statistical Generation: Three Methods Compared And Evaluated
Venue ENLG
Session Main Conference
Year 2005
Authors
  • Anja Belz (University of Brighton, Brighton UK)

Statistical NLG has largely meant n-gram mod- elling which has the considerable advantages of lending robustness to NLG systems, and of making automatic adaptation to new domains from raw cor- pora possible. On the downside, n-gram models are expensive to use as selection mechanisms and have a built-in bias towards shorter realisations. This pa- per looks at treebank-training of generators, an al- ternative method for building statistical models for NLG from raw corpora, and two different ways of using treebank-trained models during generation. Results show that the treebank-trained generators achieve improvements similar to a 2-gram gener- ator over a baseline of random selection. How- ever, the treebank-trained generators achieve this at a much lower cost than the 2-gram generator, and w...