Dynabench: Rethinking Benchmarking in NLP

TL;DR

Dynabench is an innovative platform that enables dynamic, human-in-the-loop dataset creation and model benchmarking in NLP, aiming to produce more robust and realistic evaluation benchmarks.

Contribution

It introduces a web-based platform for dynamic dataset creation and benchmarking that integrates human and model input, addressing limitations of static benchmarks.

Findings

Supports human-and-model-in-the-loop dataset creation

Addresses shortcomings of static benchmarks in NLP

Demonstrated on four initial NLP tasks

Abstract

We introduce Dynabench, an open-source platform for dynamic dataset creation and model benchmarking. Dynabench runs in a web browser and supports human-and-model-in-the-loop dataset creation: annotators seek to create examples that a target model will misclassify, but that another person will not. In this paper, we argue that Dynabench addresses a critical need in our community: contemporary models quickly achieve outstanding performance on benchmark tasks but nonetheless fail on simple challenge examples and falter in real-world scenarios. With Dynabench, dataset creation, model development, and model assessment can directly inform each other, leading to more robust and informative benchmarks. We report on four initial NLP tasks, illustrating these concepts and highlighting the promise of the platform, and address potential objections to dynamic benchmarking as a new standard for the field.