Probing for Bridging Inference in Transformer Language Models
This work addresses the problem of understanding linguistic inference capabilities in language models for NLP researchers, but it is incremental as it probes existing models without introducing new methods.
The study investigated whether pre-trained transformer language models like BERT capture bridging inference, finding that higher-layer attention heads focus on bridging relations and that models perform well on anaphora resolution tasks without fine-tuning, indicating substantial capture of this inference.
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.