START Conference Manager    

The Effect of Higher-Order Dependency Features in Discriminative Phrase-Structure Parsing

Greg Coppola and Mark Steedman

The 51st Annual Meeting of the Association for Computational Linguistics - Short Papers (ACL Short Papers 2013)
Sofia, Bulgaria, August 4-9, 2013


Abstract

We investigate the incorporation of higher-order dependency features into a cube decoding phrase-structure parser with the basic feature set of Charniak & Johnson (2005). We show that the inclusion of these features results in statistically significant gains in accuracy on a range of standard metrics. This effect is seen on both in- and out-of-domain tests. The out-of-domain results are perhaps surprising, and of special interest given that such improvements are a current goal in the field. Gains are detected whether a strong generative parser’s model score is, or is not, available to the prediction model.


START Conference Manager (V2.61.0 - Rev. 2792M)