Inducing Relational Knowledge from BERT

TL;DR

This paper investigates how well BERT captures relational knowledge by developing a method to distill such knowledge from the model using relation-specific templates and fine-tuning.

Contribution

It introduces a novel methodology for extracting and fine-tuning BERT to recognize relational knowledge beyond standard embeddings.

Findings

BERT encodes some relational knowledge that can be extracted.

The proposed method effectively identifies relation instances.

Relational knowledge extraction improves with fine-tuning on relation-specific data.

Abstract

One of the most remarkable properties of word embeddings is the fact that they capture certain types of semantic and syntactic relationships. Recently, pre-trained language models such as BERT have achieved groundbreaking results across a wide range of Natural Language Processing tasks. However, it is unclear to what extent such models capture relational knowledge beyond what is already captured by standard word embeddings. To explore this question, we propose a methodology for distilling relational knowledge from a pre-trained language model. Starting from a few seed instances of a given relation, we first use a large text corpus to find sentences that are likely to express this relation. We then use a subset of these extracted sentences as templates. Finally, we fine-tune a language model to predict whether a given word pair is likely to be an instance of some relation, when given an instantiated template for that relation as input.