Adapting Coreference Resolution Models through Active Learning
This work addresses the challenge of adapting coreference resolution models to new domains with limited data, though it is incremental as it applies existing active learning concepts to a specific task.
The paper tackled the problem of neural coreference resolution models not transferring well to low-resource domains by exploring active learning strategies, finding that labeling spans within the same document is more effective than across documents in synthetic and human experiments.
Neural coreference resolution models trained on one dataset may not transfer to new, low-resource domains. Active learning mitigates this problem by sampling a small subset of data for annotators to label. While active learning is well-defined for classification tasks, its application to coreference resolution is neither well-defined nor fully understood. This paper explores how to actively label coreference, examining sources of model uncertainty and document reading costs. We compare uncertainty sampling strategies and their advantages through thorough error analysis. In both synthetic and human experiments, labeling spans within the same document is more effective than annotating spans across documents. The findings contribute to a more realistic development of coreference resolution models.