Machine learning approaches for Chinese shallow parsers

In this paper, we present two machine-learning algorithms, namely, transformation-based error-driven learning (TEL) and memory-based learning (MBL) to improve the performance of a Chinese shallow parser. The Algorithm not only can handle nested chunking data, but also different phrase types (e.g. NP, VP, S etc.). Results show that TEL can achieve better recall rate, yet MBL is less sensitive to nesting and requires much less computation.