BED: Chinese Word Segmentation Model Based on Boundary-Enhanced Decoder

Chinese Word Segmentation (CWS) is a critical initial step in the Chinese Natural Language Processing (NLP) pipeline. Recent advancements in deep learning and pre-training language models, such as BERT and Roberta, have significantly improved CWS model performance. Nevertheless, the issue of poor performance with Out-Of-Vocabulary (OOV) words remains a challenge that warrants further exploration. Existing CWS approaches primarily focus on optimizing the model's encoder, with little attention given to enhancing the decoder's performance. This paper presents a novel solution to this gap through a Boundary-Enhanced Decoder (BED) for the CWS model. This optimized decoder brings a 0.05% improvement on Average-F1—a measure of the model's accuracy—and a 0.69% improvement on OOV Average-F1, signifying enhanced performance with OOV words. These results were obtained on four benchmark datasets using a model with a BERT encoder and softmax standard decoder. We also make our BED implementation available for further research and development.