The accuracy of Chinese parsers trained on Penn Chinese Treebank is evidently lower than that of the English parsers trained on Penn Treebank. It is plausible that the essential reason is the lack of surface syntactic constraints in Chinese. In this paper, we present evidences to show that strict deep syntactic constraints exist in Chinese sentences and such constraints cannot be effectively described with context-free phrase structure rules as in the Penn Chinese Treebank annotation; we show that such constraints may be described precisely by the idea of Sentence Structure Grammar; we introduce how to develop a broad-coverage rule-based grammar for Chinese based on this idea; we evaluated the grammar and the evaluation results show that the coverage of the current grammar is 94.2%.
展开▼