Embracing data abundance: BookTest Dataset for Reading Comprehension
This addresses the data scarcity issue in NLP for researchers and practitioners, though it is incremental as it builds on existing datasets and models.
The authors tackled the problem of limited data in reading comprehension by introducing the BookTest dataset, which is over 60 times larger than the Children's Book Test, and showed that training on this data improved model accuracy on the original test by a larger margin than architectural improvements, with an ensemble even exceeding a human baseline in one version.
There is a practically unlimited amount of natural language data available. Still, recent work in text comprehension has focused on datasets which are small relative to current computing possibilities. This article is making a case for the community to move to larger data and as a step in that direction it is proposing the BookTest, a new dataset similar to the popular Children's Book Test (CBT), however more than 60 times larger. We show that training on the new data improves the accuracy of our Attention-Sum Reader model on the original CBT test data by a much larger margin than many recent attempts to improve the model architecture. On one version of the dataset our ensemble even exceeds the human baseline provided by Facebook. We then show in our own human study that there is still space for further improvement.