On Generalization in Coreference Resolution
This work addresses the domain generalization issue in coreference resolution for NLP researchers, offering a new benchmark and improved models, though it is incremental as it builds on existing methods with data mixing.
The paper tackled the problem of poor generalization in coreference resolution models across unseen domains by consolidating 8 datasets and proposing a joint training method with data augmentation and sampling. The result showed that joint training improved overall performance in zero-shot settings, leading to better generalization and multiple new state-of-the-art results.
While coreference resolution is defined independently of dataset domain, most models for performing coreference resolution do not transfer well to unseen domains. We consolidate a set of 8 coreference resolution datasets targeting different domains to evaluate the off-the-shelf performance of models. We then mix three datasets for training; even though their domain, annotation guidelines, and metadata differ, we propose a method for jointly training a single model on this heterogeneous data mixture by using data augmentation to account for annotation differences and sampling to balance the data quantities. We find that in a zero-shot setting, models trained on a single dataset transfer poorly while joint training yields improved overall performance, leading to better generalization in coreference resolution models. This work contributes a new benchmark for robust coreference resolution and multiple new state-of-the-art results.