- Abstract: Machine reading comprehension is a key part of natural language understanding. Due to wide range of applications, machine reading comprehension has attracted considerable in- terest from both commercial and academic en- tities. Recently, deep neural network based models have been able to achieve near human performance on certain types of reading com- prehension datasets (Rajpurkar et al., 2016; Devlin et al., 2018; Hermann et al., 2015; Seo et al., 2016). Newer and much harder datasets have been proposed that require more sophisti- cated reasoning capabilities. One such dataset was proposed by Khashabi et al.(2018) called MultiRC. In this work, we propose a multi- stage approach which significantly improves the previous state of the art results on MultiRC (Khashabi et al., 2018).
- Keywords: question answering