Adversarial NLI: A New Benchmark for Natural Language Understanding

TL;DR

This paper presents a new adversarially collected NLI benchmark dataset that challenges current models, reveals their weaknesses, and can evolve over time to continually advance natural language understanding.

Contribution

It introduces a large-scale, adversarially collected NLI dataset that improves model performance and highlights current limitations, with a novel, adaptable data collection method.

Findings

Models trained on the dataset achieve state-of-the-art results.

The new test set is more challenging for existing models.

Non-expert annotators effectively identify model weaknesses.

Abstract

We introduce a new large-scale NLI benchmark dataset, collected via an iterative, adversarial human-and-model-in-the-loop procedure. We show that training models on this new dataset leads to state-of-the-art performance on a variety of popular NLI benchmarks, while posing a more difficult challenge with its new test set. Our analysis sheds light on the shortcomings of current state-of-the-art models, and shows that non-expert annotators are successful at finding their weaknesses. The data collection method can be applied in a never-ending learning scenario, becoming a moving target for NLU, rather than a static benchmark that will quickly saturate.