Probing for Bridging Inference in Transformer Language Models

TL;DR

This paper investigates how transformer language models, especially BERT, inherently understand bridging inference, revealing that higher layers and specific attention heads focus on bridging relations, and that models can perform bridging anaphora resolution without fine-tuning.

Contribution

It provides the first detailed analysis of bridging inference in transformer models and introduces a masked token prediction approach to evaluate this capability without additional training.

Findings

Higher layers focus on bridging relations in BERT.

Specific attention heads consistently target bridging.

Pre-trained models can resolve bridging anaphora without fine-tuning.

Abstract

We probe pre-trained transformer language models for bridging inference. We first investigate individual attention heads in BERT and observe that attention heads at higher layers prominently focus on bridging relations in-comparison with the lower and middle layers, also, few specific attention heads concentrate consistently on bridging. More importantly, we consider language models as a whole in our second approach where bridging anaphora resolution is formulated as a masked token prediction task (Of-Cloze test). Our formulation produces optimistic results without any fine-tuning, which indicates that pre-trained language models substantially capture bridging inference. Our further investigation shows that the distance between anaphor-antecedent and the context provided to language models play an important role in the inference.