Choice-Driven Contextual Reasoning for Commonsense Question Answering

Wenqing Deng, Zhe Wang, Kewen Wang, Xiaowang Zhang, Zhiyong Feng

Abstract: The task of question answering is to find the most appropriate answer for an input question in natural language from a given custom knowledge base of information. While the performance of question answering systems has been significantly improved, they still struggle to answer questions that require commonsense reasoning. To capture common sense beyond associations, a challenging dataset CommonsenseQA for commonsense question answering is proposed. As a result, several models have been developed for tackling this challenge. But existing approaches are still limited in handling contextual representation and reasoning. In this paper, we propose a model for commonsense question answering by implementing a form of choice-driven contextual reasoning through novel encoding strategies and choice differentiation mechanisms. We have conducted experiments on major baselines for commonsense question answering and our experimental results show that the proposed model significantly outperforms strong baselines.