We present a novel representation of parse trees aslists of paths (leaf projection paths) from leaves tothe top level of the tree. This representation allowsus to achieve significantly higher accuracy in thetask of HPSG parse selection than standard models,and makes the application of string kernels natural.We define tree kernels via string kernels on projectionpaths and explore their performance in the contextof parse disambiguation. We apply SVM rankingmodels and achieve an exact sentence accuracyof 85.40% on the Redwoods corpus.
展开▼