Dong Zhang, Zheng Hu, Shoushan Li, Hanqian Wu, Qiaoming Zhu, Guodong Zhou. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 2: Short Papers). 2021.
As one of the basic tasks in natural language processing (NLP), Chinese word segmentation (CWS) is a necessary and key preprocessing step for many natural language processing tasks. The performance of CWS directly affects the final performance of subsequent tasks. Traditional word segmentation approaches mainly use conditional random field (CRF) decoder and BIO tags for sequence labeling, and the recent mainstream approaches normally employ the pre-trained bidirectional encoder representations from transformers (BERT) model as text encoder. This greatly affects the speed of word segmentation. Different from the existing approaches, this paper proposes a Chinese word segmentation approach based on word boundary classification, which utilizes a lite BERT (ALBERT) model as the text encoder. This approach directly determines whether the position between characters is the boundary between two potential words without using CRF decoding. Experimental results show that our proposed approach can achieve significant improvements, and significantly reduces the training and testing time of CWS.