Materialized Knowledge Bases from Commonsense Transformers

TL;DR

This paper introduces a publicly available resource of commonsense knowledge generated from pre-trained language models, enabling new analyses and applications in the field.

Contribution

It provides the first materialized commonsense knowledge base from transformers, along with detailed evaluation and discussion of its potential uses.

Findings

High precision and recall in generated knowledge

Identification of common problem cases

Enabling off-the-shelf use of commonsense knowledge

Abstract

Starting from the COMET methodology by Bosselut et al. (2019), generating commonsense knowledge directly from pre-trained language models has recently received significant attention. Surprisingly, up to now no materialized resource of commonsense knowledge generated this way is publicly available. This paper fills this gap, and uses the materialized resources to perform a detailed analysis of the potential of this approach in terms of precision and recall. Furthermore, we identify common problem cases, and outline use cases enabled by materialized resources. We posit that the availability of these resources is important for the advancement of the field, as it enables an off-the-shelf-use of the resulting knowledge, as well as further analyses on its strengths and weaknesses.