Abstract: Named Entity Recognition (NER) is designed to identify and classify proper nouns in text.Training data for supervised learning are usually manually annotated,and it is difficult to obtain large-scale annotated data due to time-consuming and labor-intensive.In order to solve the problem of data sparseness caused by the lack of large-scale annotation corpus and the problem of polysemy of charembeddingin the Chinese NER task,this paper uses contextualized char embeddings which is pre-trained on large-scale unlabeled data to improve the performance of the Chinese NER model.Furthermore,to solve the problem of out-of-vocabulary words in named entity recognition,this paper proposes a Chinese NER system based on word language model.We use the contextualized char embeddings of generated by the language model as the input of the NER model to capture different mea-nings of Chinese characters in different contexts.In this paper,we conduct experiments on six Chinese NER datasets.The experimental results show that the proposed model can improve the performance and the average F1 improves by 4.95%.In addition,this paper further analyzes the experimental results and finds that the proposed model can achieve better results on OOV entities,and it has good performance for some special types of Chinese entity recognition.

Key words: Contextualized char vector, Language model, Named entity recognition