Teaching Machines to Read and Comprehend

TL;DR

This paper introduces a new methodology for creating large-scale supervised reading comprehension datasets, enabling the development of neural networks that can read and answer questions about real documents with minimal prior language knowledge.

Contribution

It presents a novel methodology for generating large-scale datasets and develops attention-based neural networks capable of understanding and answering questions on real documents.

Findings

Successful training of neural networks on new large-scale datasets

Neural models can answer complex questions with minimal language prior knowledge

Improved performance over previous methods in machine reading comprehension

Abstract

Teaching machines to read natural language documents remains an elusive challenge. Machine reading systems can be tested on their ability to answer questions posed on the contents of documents that they have seen, but until now large scale training and test datasets have been missing for this type of evaluation. In this work we define a new methodology that resolves this bottleneck and provides large scale supervised reading comprehension data. This allows us to develop a class of attention based deep neural networks that learn to read real documents and answer complex questions with minimal prior knowledge of language structure.