Constructing Datasets for Multi-hop Reading Comprehension Across Documents (arXiv)

Question answering (QA) has seen many improvements in recent years, particularly fuelled by new datasets such as SQuAD. Existing datasets, however, focus on single-hop reading comprehension, i.e. extracting an answer to a question from a given paragraph. Welbl et al. introduce two new datasets for multi-hop reading comprehension, which is much closer the real-world task of open-domain question answering. The datasets require models to first identify relevant documents among a number of candidate documents and then determine the correct answer.


Want to receive more content like this in your inbox?