Rethinking Chinese Word Segmentation: Tokenization, Character Classification, or Wordbreak Identification