Unsupervised Segmentation Helps Supervised Learning of Character Tagging for Word Segmentation and Named Entity Recognition

This paper describes a novel character tagging approach to Chinese word segmentation and named entity recognition (NER) for our participation in Bakeoff-4.1 It integrates unsupervised segmentation and conditional random fields (CRFs) learning successfully, using similar character tags and feature templates for both word segmentation and NER. It ranks at the top in all closed tests of word segmentation and gives promising results for all closed and open NER tasks in the Bakeoff. Tag set selection and unsupervised segmentation play a critical role in this success.