Active Learning for Coreference Resolution
Timothy A. Miller, Dmitriy Dligach, Guergana Savova · North American Chapter of the Association for Computational Linguistics · 2012
Active learning can lower the cost of annotation for some natural language processing tasks by using a classifier to select informative instances to send to human annotators. It has worked well in cases where the training instances are selected one at a time and require minimal context for annotation. However, coreference annotations often require some context and the traditional active learning approach may not be feasible. In this work we explore various active learning methods for coreference resolution that fit more realistically into coreference annotation workflows.