Every Answer Matters: Evaluating Commonsense with Probabilistic Measures

TL;DR

This paper introduces a new generative task and probabilistic evaluation method to better assess the inherently probabilistic nature of commonsense reasoning in language models, addressing limitations of existing multiple-choice benchmarks.

Contribution

It presents the commonsense frame completion (CFC) task and a probabilistic evaluation approach that correlates with human judgments, providing a more accurate measure of machine commonsense.

Findings

Humans outperform language models on CFC.

Probabilistic evaluation correlates well with human judgments.

Existing benchmarks do not capture the probabilistic aspect of commonsense.

Abstract

Large language models have demonstrated impressive performance on commonsense tasks; however, these tasks are often posed as multiple-choice questions, allowing models to exploit systematic biases. Commonsense is also inherently probabilistic with multiple correct answers. The purpose of "boiling water" could be making tea and cooking, but it also could be killing germs. Existing tasks do not capture the probabilistic nature of common sense. To this end, we present commonsense frame completion (CFC), a new generative task that evaluates common sense via multiple open-ended generations. We also propose a method of probabilistic evaluation that strongly correlates with human judgments. Humans drastically outperform strong language model baselines on our dataset, indicating this approach is both a challenging and useful evaluation of machine common sense.