Multiple Attention Networks with Temporal Convolution for Machine Reading Comprehension
Jiabao Guo, Gang Liu, Caiquan Xiong · 2019
Machine reading comprehension (MRC) is an important branch of natural language processing (NLP). In this paper, a new neural structure called multiple attention network with temporal convolution (MATC-net) is proposed for reading comprehension style question answering, which aims to answer questions from a given passage. The main contribution and originality of MATC-net are as follows: 1) a novel deep bidirectional language model is employed to produce better word-level embedding vectors. 2) the convoluteional architectures and bidirectional long short-term memory are combined for the comprehension text encoding process. 3) the proposed multiple attention mechanisms are designed to capture the complete information and exploit it in its counterpart layer by layer. Experimental verification is conducted on Stanford Question Answering Dataset (SQuAD). The results clearly show that our model achieves improvements on previous public reading comprehension.