Inducing Head-Driven PCFGs with Latent Heads: Refining a Tree-Bank Grammar for Parsing

摘要: Although state-of-the-art parsers for natural language are lexicalized, it was recently shown that an accurate unlexicalized parser the Penn tree-bank can be simply read off a manually refined tree-bank. While lexicalized often suffer from sparse data, manual mark-up is costly and largely based on individual linguistic intuition. Thus, across domains, languages, annotations, fundamental question arises: Is possible to automatically induce without resorting full lexicalization? In this paper, we show how probabilistic with latent head information simple principles. Our has performance of 85.1% (LP/LR F1), which as good early ones. This remarkable since induction grammars in general hard task.