Co-regularizing character-based and word-based models for semi-supervised Chinese word segmentation.