Higher-order Coreference Resolution with Coarse-to-fine Inference

TL;DR

This paper presents a differentiable higher-order inference method for coreference resolution that iteratively refines span representations using a coarse-to-fine approach, improving accuracy and efficiency.

Contribution

It introduces a novel fully differentiable approximation for higher-order inference with a coarse-to-fine pruning strategy, enhancing coreference resolution performance.

Findings

Significant accuracy improvement on the English OntoNotes benchmark

Enhanced computational efficiency over previous span-ranking models

Effective iterative refinement of span representations

Abstract

We introduce a fully differentiable approximation to higher-order inference for coreference resolution. Our approach uses the antecedent distribution from a span-ranking architecture as an attention mechanism to iteratively refine span representations. This enables the model to softly consider multiple hops in the predicted clusters. To alleviate the computational cost of this iterative process, we introduce a coarse-to-fine approach that incorporates a less accurate but more efficient bilinear factor, enabling more aggressive pruning without hurting accuracy. Compared to the existing state-of-the-art span-ranking approach, our model significantly improves accuracy on the English OntoNotes benchmark, while being far more computationally efficient.