featured.png

We present multi-hop question answering datasets from two different domains, designed to enabe models to combine disjoint pieces of textual evidence to extend the scope of machine comprehension methods.