第一遍：

亮点

1、通过引用外部数据，整合到预训练模型中，通过实验证实其有效

• First, we identify concepts in question and answer options and link these potentially ambiguous concepts to an open domain resource that provides unstructured background information relevant to the concepts and used to enrich the original reference corpus
• In comparison to previous work (e.g., (Yadav et al., 2019)), we perform informa-tion retrieval based on the enriched corpus instead of the original one to form a document for answering a question。
• Second, we increase the amount of training data by appending additional in-domain subject-area QA datasets

2、但当添加的数据集相比赛题更加“生僻”时，其准确率会有所下降

• performance degrades when the added data exhibit a higher level of difficulty than the original training data