Word Boundary Decision with CRF for Chinese Word Segmentation

Chinese word segmentation systems necessarily perform both accurately and quickly for real applications. In this paper, we study on word boundary decision (WBD) approach for Chinese word segmentation and implement it as a 2-tag character tagging with conditional random filed (CRF). With a help of tag transition features, WBD with CRF segmentation approach can achieve comparative performances compared to 4-tag character tagging approach (represents the state-of-the-art segmentation approach). But it requires only about half training time and memory space as much as 4-tag character tagging approach. These results encourage that WBD segmentation approach is a good choice for real Chinese word segmentation systems.