Prior Knowledge Driven Domain Adaptation

The performance of a natural language system trained on one domain often drops significantly when testing on another domain. Therefore, the problem of domain adaptation remains one of the most important natural language processing challenges. While many different domain adaptation frameworks have been proposed, they have ignored one natural resource – the prior knowledge on the new domain. In this paper, we propose a new adaptation framework called Prior knowledge Driven Adaptation (PDA), which takes advantage of the knowledge on the target domain to adapt the model. We empirically study the effects of incorporating prior knowledge in different ways. On the task of part-of-speech tagging, we show that prior knowledge results in 42% error reduction when adapting from news text to biomedical text. On the task of semantic role labeling, when adapting from one news domain to another news domain, prior knowledge gives 25% error reduction for instances of be verbs (unseen in training domain) and 9% error reduction over instances of all verbs.