Distinguish Before Answer: Generating Contrastive Explanation as Knowledge for Commonsense Question Answering

TL;DR

The paper introduces CPACE, a model that generates contrastive explanations from symbolic knowledge to improve commonsense question answering, achieving state-of-the-art results on multiple datasets.

Contribution

It proposes a novel contrastive explanation generation method that enhances knowledge distinguishment and interpretability for QA tasks.

Findings

Achieves new SOTA on CSQA with 89.8% accuracy.

Significant improvements on QASC and OBQA datasets.

Effectively utilizes generated explanations for downstream QA enhancement.

Abstract

Existing knowledge-enhanced methods have achieved remarkable results in certain QA tasks via obtaining diverse knowledge from different knowledge bases. However, limited by the properties of retrieved knowledge, they still have trouble benefiting from both the knowledge relevance and distinguishment simultaneously. To address the challenge, we propose CPACE, a Concept-centric Prompt-bAsed Contrastive Explanation Generation model, which aims to convert obtained symbolic knowledge into a contrastive explanation for better distinguishing the differences among given candidates. Firstly, following previous works, we retrieve different types of symbolic knowledge with a concept-centric knowledge extraction module. After that, we generate corresponding contrastive explanations using acquired symbolic knowledge and explanation prompts as guidance for better modeling the knowledge distinguishment and interpretability. Finally, we regard the generated contrastive explanation as external knowledge for downstream task enhancement. We conduct a series of experiments on three widely-used question-answering datasets: CSQA, QASC, and OBQA. Experimental results demonstrate that with the help of generated contrastive explanation, our CPACE model achieves new SOTA on CSQA (89.8% on the testing set, 0.9% higher than human performance), and gains impressive improvement on QASC and OBQA (4.2% and 3.5%, respectively).