Hermann et al. (2015) created two useful datasets using news articles for the purpose of Q&A research. Dataset contains 90k and 197k documents each, and each document is accompanied, on average, by 4 questions. Each question is a sentence with one missing word or phrase that can be found from the accompanying documents.

QNLI Dataset
DOWNLOAD QNLI
https://data.deepai.org/QNLIv2.zip