Improving Question Answering with External Knowledge

Prior background knowledge is essential for human reading and understanding. In this work, we investigate how to leverage external knowledge to improve question answering. We primarily focus on multiple-choice question answering tasks that require external knowledge to answer questions. We investigate the effects of utilizing external in-domain multiple-choice question answering datasets and enriching the reference corpus by external out-domain corpora (i.e., Wikipedia articles). Experimental results demonstrate the effectiveness of external knowledge on two challenging multiple-choice question answering tasks: ARC and OpenBookQA.