Domain Adaptation with Active Learning for Coreference Resolution

In the literature, most prior work on coreference resolution centered on the newswire domain. Although a coreference resolution system trained on the newswire domain performs well on newswire texts, there is a huge performance drop when it is applied to the biomedical domain. In this paper, we present an approach integrating domain adaptation with active learning to adapt coreference resolution from the newswire domain to the biomedical domain. We explore the effect of domain adaptation, active learning, and target domain instance weighting for coreference resolution. Experimental results show that domain adaptation with active learning and target domain instance weighting achieves performance on MEDLINE abstracts similar to a system trained on coreference annotation of only target domain training instances, but with a greatly reduced number of target domain training instances that we need to annotate.