Simplify the Usage of Lexicon in Chinese NER

• DOI: 10.18653/v1/2020.acl-main.528
• Publication Date: 2020-07-29T14:14:43Z
• Authors: Ruotian Ma; Minlong Peng; Qi Zhang; Zhongyu Wei; Xuanjing Huang
• Keywords: Named Entity Recognition; Benchmark (surveying); Representation

Recently, many works have tried to augment the performance of Chinese named entity recognition (NER) using word lexicons. As a representative, Lattice-LSTM has achieved new benchmark results on several public NER datasets. However, complex model architecture. This limits its application in industrial areas where real-time responses are needed. In this work, we propose simple but effective method for incorporating lexicon into character representations. avoids designing complicated sequence modeling architecture, and any neural model, it requires only subtle adjustment representation layer introduce information. Experimental studies four datasets show that our achieves an inference speed up 6.15 times faster than those state-of-the-art methods, along with better performance. The experimental also proposed can be easily incorporated pre-trained models like BERT.