Alleviating the Knowledge-Language Inconsistency: A Study for Deep Commonsense Knowledge

TL;DR

This paper identifies a form of deep commonsense knowledge represented in natural language sentences that traditional triple-based methods struggle to capture, and proposes a novel sentence-based mining approach to improve extraction.

Contribution

It introduces the concept of deep commonsense knowledge, highlights limitations of existing triple-based methods, and proposes a new sentence-based mining method that enhances knowledge extraction.

Findings

Deep commonsense knowledge constitutes a significant portion of overall commonsense knowledge.

Conventional triple-based methods are ineffective at capturing deep commonsense knowledge.

The proposed sentence-based mining method significantly improves extraction performance.

Abstract

Knowledge facts are typically represented by relational triples, while we observe that some commonsense facts are represented by the triples whose forms are inconsistent with the expression of language. This inconsistency puts forward a challenge for pre-trained language models to deal with these commonsense knowledge facts. In this paper, we term such knowledge as deep commonsense knowledge and conduct extensive exploratory experiments on it. We show that deep commonsense knowledge occupies a significant part of commonsense knowledge while conventional methods fail to capture it effectively. We further propose a novel method to mine the deep commonsense knowledge distributed in sentences, alleviating the reliance of conventional methods on the triple representation form of knowledge. Experiments demonstrate that the proposal significantly improves the performance in mining deep commonsense knowledge.