Named Entity Recognition in Classical Chinese by Lexicon Enhancement

The significant differences between classical Chinese and modern Chinese make the NER methods that are applicable to modern Chinese not applicable to classical Chinese. In this paper, we propose a NER model for classical Chinese based on a lexicon-enhanced BERT model, where lexical features are integrated into the encoder layer. Experiment results show that lexicon enhancement on the classical Chinese BERT model can improve the performance of NER. Additionally, the effectiveness of fusing lexical information depends on the quality of word embeddings.