Improving Chinese Dependency Parsing with Auto-extracted Dependency Triples

To solve the data sparseness problem in dependency parsing, most previous studies used features extracted from large-scale auto-parsed data. Unlike previous work, we propose a novel approach to improve dependency parsing with dependency triples (DT) extracted by self-disambiguating patterns (SDP). The use of SDP makes it possible to avoid the dependency on a baseline parser and explore the influence of different types of DTs one by one. Experiments show that, when DT features are integrated into a maximum spanning tree (MST) dependency parser, the new parser improves significantly over the baseline MST parser. Comparative results also show that DTs with dependency relation labels perform much better than DTs without dependency relation label.