Mnemonic Reader: Machine Comprehension with Iterative Aligning and Multi-hop Answer Pointing

Recently, several end-to-end neural models have been proposed for machine comprehension (MC) tasks. Most of these models only capture interactions between the context and the query, and utilize "one-shot prediction" to point the boundary of answer, failing to fully understand the context and the query. In this paper, we introduce Mnemonic Reader for MC tasks, an end-to-end neural network which aims to tackle the above problem in two aspects. Firstly, we propose an iterative aligning mechanism which not only captures interactions between the context and the query but also models interactions among the context itself to obtain a fully-aware context representation. Second, we use a multi-hop answer pointer which allows the network to continue refining the predicted answer span. Extensive experiments on TriviaQA and SQuAD datasets show that our model obtains state-of-the-art results.