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

Abstract

Although state-of-the-art parsers for natural language are lexicalized, it was recently shown that an accurate unlexicalized parser for the Penn tree-bank can be simply read off a manually refined tree-bank. While lexicalized parsers often suffer from sparse data, manual mark-up is costly and largely based on individual linguistic intuition. Thus, across domains, languages, and tree-bank annotations, a fundamental question arises: Is it possible to automatically induce an accurate parser from a tree-bank without resorting to full lexicalization? In this paper, we show how to induce a probabilistic parser with latent head information from simple linguistic principles. Our parser has a performance of 85.1% (LP/LR F_1), which is as good as that of early lexicalized ones. This is remarkable since the induction of probabilistic grammars is in general a hard task.