A Chinese Chunk-based Parser

The technique of multi-level chunking has been applied to full sentence parsing in a number of previous studies. Instead of using the popular IOB chunking technique, this paper explores a novel method of locating chunks by identifying their boundaries. We investigated the use of a machine learning algorithm to classify the boundaries using the part-of-speech (POS) of words and their associated information-theoretic measures. A phrase recognizer is developed to predict the syntactic class of the identified chunks. While the training is conducted using a partial version of a treebank with less than 360,000 POS-tagged tokens, our system achieves a performance of 80.46 % labelled recall, 85.70 % labelled precision, and 83.00 % F-score in the testing of 93,000 POS-tagged tokens, without any assumption of prior knowledge or statistical measures from any other resources