On Generalization in Coreference Resolution

TL;DR

This paper investigates the generalization of coreference resolution models across different domains, proposing a joint training method on heterogeneous datasets that improves zero-shot performance and sets new benchmarks.

Contribution

It introduces a novel joint training approach with data augmentation for heterogeneous datasets, enhancing model generalization in coreference resolution.

Findings

Joint training improves zero-shot transfer performance.

Data augmentation helps handle annotation differences.

Achieves new state-of-the-art results on a robust coreference benchmark.

Abstract

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.