Chinese deterministic dependency analysis with consideration of long-distance dependency
Huiwei Zhou, Degen Huang, Yang Yage · International Conference on Intelligent Systems and Control · 2007
According to Chinese syntax, implement a deterministic Chinese dependency analyzer based on an improved Nivre's algorithm which considers long-distance dependency. It is difficult to parse long-distance dependency with conventional deterministic dependency analysis methods. The proposed method parses a sentence deterministically without ignoring long-distance dependency. In addition, we also construct a root node finder to divide the sentence into two sub-sentences. Support Vector Machines are applied to identify Chinese dependency structure. We compare the performance of two sorts of classifiers -- Support Vector Machines and Preference Learning in root node finding. Experiments using the Harbin University of Technology Corpus show that the method outperforms previous system by 6.46% accuracy. The dependency accuracy achieves 79.44% even with small training data (4000 sentences).