DROP: A Reading Comprehension Benchmark Requiring Discrete Reasoning Over Paragraphs

TL;DR

DROP is a challenging reading comprehension benchmark that requires discrete reasoning over paragraphs, revealing significant gaps in current models' understanding compared to human performance.

Contribution

This paper introduces DROP, a new benchmark dataset emphasizing discrete reasoning, and demonstrates the limitations of existing models while proposing a combined approach for improved performance.

Findings

Current models achieve only 32.7% F1 on DROP

Humans reach 96.0% F1 on the dataset

A new combined model improves F1 to 47.0%

Abstract

Reading comprehension has recently seen rapid progress, with systems matching humans on the most popular datasets for the task. However, a large body of work has highlighted the brittleness of these systems, showing that there is much work left to be done. We introduce a new English reading comprehension benchmark, DROP, which requires Discrete Reasoning Over the content of Paragraphs. In this crowdsourced, adversarially-created, 96k-question benchmark, a system must resolve references in a question, perhaps to multiple input positions, and perform discrete operations over them (such as addition, counting, or sorting). These operations require a much more comprehensive understanding of the content of paragraphs than what was necessary for prior datasets. We apply state-of-the-art methods from both the reading comprehension and semantic parsing literature on this dataset and show that the best systems only achieve 32.7% F1 on our generalized accuracy metric, while expert human performance is 96.0%. We additionally present a new model that combines reading comprehension methods with simple numerical reasoning to achieve 47.0% F1.