DUMA: Reading Comprehension With Transposition Thinking

Multi-choice Machine Reading Comprehension (MRC) requires model to decide the correct answer from a set of options when given passage and question. Thus in addition powerful Pre-trained Language Model (PrLM) as encoder, multi-choice MRC especially relies on matching network design which is supposed effectively capture relationships among triplet passage, question answers. While newer more PrLMs have shown their mightiness even without support network, we propose new DUal Multi-head Co-Attention (DUMA) model, inspired by human's transposition thinking process solving problem: respectively considering each other's focus standpoint The proposed DUMA has been effective capable generally promoting PrLMs. Our method evaluated two benchmark tasks, DREAM RACE, showing that terms PrLMs, can still boost reach state-of-the-art performance.