Portal de Periódicos da CAPES

Is Local Window Essential for Neural Network Based Chinese Word Segmentation?

2016; Springer Science+Business Media; Linguagem: Inglês

10.1007/978-3-319-47674-2_37

1611-3349

Jinchao Zhang, Fandong Meng, Mingxuan Wang, Daqi Zheng, Wenbin Jiang, Qun Liu,

Speech Recognition and Synthesis

Neural network based Chinese Word Segmentation (CWS) approaches can bypass the burdensome feature engineering comparing with the conventional ones. All previous neural network based approaches rely on a local window in character sequence labelling process. It can hardly exploit the outer context and may preserve indifferent inner context. Moreover, the size of local window is a toilsome manual-tuned hyper-parameter that has significant influence on model performance. We are wondering if the local window can be discarded in neural network based CWS. In this paper, we present a window-free Bi-directional Long Short-term Memory (Bi-LSTM) neural network based Chinese word segmentation model. The model takes the whole sentence under consideration to generate reasonable word sequence. The experiments show that the Bi-LSTM can learn sufficient context for CWS without the local window.