Data-Oriented Parsing and the Penn Chinese Treebank

We present an investigation into parsing the Penn Chinese Treebank using a DataOriented Parsing (DOP) approach. DOP comprises an experience-based approach to natural language parsing. Most published research in the DOP framework uses PStrees as its representation schema. Drawbacks of the DOP approach centre around issues of efficiency. We incorporate recent advances in DOP parsing techniques into a novel DOP parser which generates a compact representation of all subtrees which can be derived from any full parse tree. We compare our work to previous work on parsing the Penn Chinese Treebank, and provide both a quantitative and qualitative evaluation. While our results in terms of Precision and Recall are slightly below those published in related research, our approach requires no manual encoding of head rules, nor is a development phase per se necessary. We also note that certain constructions which were problematic in this previous work can be handled correctly by our DOP parser. Finally, we observe that the ‘DOP Hypothesis’ is confirmed for parsing the Penn Chinese Treebank.