Using Linguistic Features to Improve the Generalization Capability of Neural Coreference Resolvers

TL;DR

This paper explores how incorporating specific linguistic features into neural coreference resolvers enhances their ability to generalize across different domains, leading to state-of-the-art out-of-domain performance.

Contribution

The study demonstrates that selecting informative linguistic features significantly improves the generalization of neural coreference resolvers beyond the training domain.

Findings

Incorporating linguistic features slightly improves generalization.

Using informative feature subsets greatly enhances out-of-domain performance.

Achieves state-of-the-art results on WikiCoref without domain-specific training.

Abstract

Coreference resolution is an intermediate step for text understanding. It is used in tasks and domains for which we do not necessarily have coreference annotated corpora. Therefore, generalization is of special importance for coreference resolution. However, while recent coreference resolvers have notable improvements on the CoNLL dataset, they struggle to generalize properly to new domains or datasets. In this paper, we investigate the role of linguistic features in building more generalizable coreference resolvers. We show that generalization improves only slightly by merely using a set of additional linguistic features. However, employing features and subsets of their values that are informative for coreference resolution, considerably improves generalization. Thanks to better generalization, our system achieves state-of-the-art results in out-of-domain evaluations, e.g., on WikiCoref, our system, which is trained on CoNLL, achieves on-par performance with a system designed for this dataset.