Abstract
We propose three improvements to address the drawbacks of state-of-the-art transition-based constituent parsers. First, to resolve the error propagation problem of the traditional pipeline approach, we incorporate POS tagging into the syntactic parsing process. Second, to alleviate the negative influence of size differences among competing action sequences, we align parser states during beam-search decoding. Third, to enhance the power of parsing models, we enlarge the feature set with non-local features and semi-supervised word cluster features. Experimental results show that these modifications improve parsing performance significantly. Evaluated on the Chinese TreeBank (CTB), our final performance reaches 86.3% (F1) when trained on CTB 5.1, and 87.1% when trained on CTB 6.0, and these results outperform all state-of-the-art parsers.