Recently, neural network models for natural language processing tasks havebeen increasingly focused on for their ability of alleviating the burden ofmanual feature engineering. However, the previous neural models cannot extractthe complicated feature compositions as the traditional methods with discretefeatures. In this work, we propose a feature-enriched neural model for jointChinese word segmentation and part-of-speech tagging task. Specifically, tosimulate the feature templates of traditional discrete feature based models, weuse different filters to model the complex compositional features withconvolutional and pooling layer, and then utilize long distance dependencyinformation with recurrent layer. Experimental results on five differentdatasets show the effectiveness of our proposed model.
展开▼
