Unsupervised Domain Adaptation for Joint Segmentation and POS-Tagging
Yang Liu, Yue Zhang · 2012
Sophisticated models have been developed for joint word segmentation and part-of-speech tagging, with increasing accuracies reported on the Chinese Treebank data. These systems, which rely on supervised learning, typically perform worse on texts from a different domain, for which little annotation is available. We consider self-training and character clustering for domain adaptation. Both methods use only unannotated target-domain data, and are relatively straightforward to implement upon a baseline supervised system. Our results show that both methods can effectively improve target-domain performance. In addition, a combination of the two orthogonal methods leads to further improvement.