In this paper, we propose a novel model and its training strategy for the CRF.