Incorporating Statistical Information of Lexical Dependency into a Rule-Based Parser

This paper presents a method to incorporate statistical information into a rulebased parser to resolve syntactic ambiguities. We extract the statistical information from the Penn Treebank, and apply the information to the rule-based parser. For the extraction of the statistical information the tag conversion is needed because of the disagreement of the tags and the bracketing style. We will show the effect of the tag conversion with experiments. The final result shows about 7% error rate reduction in the dependency evaluation. We will also show how much each type of statistical information affects the parsing performance.