Natural Language Comprehension with the EpiReader
Adam Trischler, Zheng Ye, Xingdi Yuan, Kaheer Suleman
TL;DR
The paper presents EpiReader, a two-stage, end-to-end neural architecture for machine reading comprehension that first extracts a small set of candidate answers and then tests hypotheses against the text using textual entailment. By explicitly separating candidate selection from hypothesis testing and training them jointly, EpiReader achieves state-of-the-art results on the CNN/Daily Mail and CBT datasets. The approach highlights the value of explicit hypothesis testing in neural QA and demonstrates that a reasoner can correct errors made by a comparatively lightweight extractor. This work suggests a fruitful direction of combining fast candidate generation with deeper entailment-based verification for robust machine comprehension.
Abstract
We present the EpiReader, a novel model for machine comprehension of text. Machine comprehension of unstructured, real-world text is a major research goal for natural language processing. Current tests of machine comprehension pose questions whose answers can be inferred from some supporting text, and evaluate a model's response to the questions. The EpiReader is an end-to-end neural model comprising two components: the first component proposes a small set of candidate answers after comparing a question to its supporting text, and the second component formulates hypotheses using the proposed candidates and the question, then reranks the hypotheses based on their estimated concordance with the supporting text. We present experiments demonstrating that the EpiReader sets a new state-of-the-art on the CNN and Children's Book Test machine comprehension benchmarks, outperforming previous neural models by a significant margin.