Text this: Interpreting black-box models through sufficient input subsets