The machine reading comprehension(MRC)task challenges the machine's ability to understand natural language by asking the machine to answer questions in a given context.To improve the accuracy of answer extraction involving the crossing of punctuation sentences and long-distance correlation of answer elements,clue elements and question elements,this paper proposes to model the long-distance relationship between punctuation sentences,and complement the missing components by shared structure.A machine reading comprehension model is implemented by integrating the Naming-Telling structure information.The experimental results on the public data set CMRC2018 show that the proposed method achieves an increase of 2.4％in F1-value and 6％in EM value compared with the baseline model.
machine reading comprehensionnaming-telling structureattentionpretraining language model
维普
万方数据
