Machine Comprehension Using Match-LSTM and Answer Pointer

TLDR

Machine comprehension is a key NLP problem, and the Stanford Question Answering Dataset (SQuAD) offers a large, variable‑length, real‑question testbed that challenges algorithms beyond small candidate sets. The authors aim to develop an end‑to‑end neural architecture for machine comprehension using two variants of Pointer Net. The architecture combines match‑LSTM with Pointer Net, employing two variants that constrain output tokens to the input sequence. Both models significantly outperform prior best results on SQuAD, surpassing Rajpurkar et al.’s logistic regression baseline.

Abstract

Machine comprehension of text is an important problem in natural language processing. A recently released dataset, the Stanford Question Answering Dataset (SQuAD), offers a large number of real questions and their answers created by humans through crowdsourcing. SQuAD provides a challenging testbed for evaluating machine comprehension algorithms, partly because compared with previous datasets, in SQuAD the answers do not come from a small set of candidate answers and they have variable lengths. We propose an end-to-end neural architecture for the task. The architecture is based on match-LSTM, a model we proposed previously for textual entailment, and Pointer Net, a sequence-to-sequence model proposed by Vinyals et al.(2015) to constrain the output tokens to be from the input sequences. We propose two ways of using Pointer Net for our task. Our experiments show that both of our two models substantially outperform the best results obtained by Rajpurkar et al.(2016) using logistic regression and manually crafted features.