Most previous approaches to chinese word segmentation formalize this problem
as a character-based sequence labeling task where only contextual information
within fixed sized local windows and simple interactions between adjacent tags
can be captured. In this paper, we propose a novel n