COM2SENSE: A Commonsense Reasoning Benchmark with Complementary Sentences

TL;DR

COM2SENSE introduces a new challenging benchmark dataset with complementary sentence pairs to evaluate and analyze AI's commonsense reasoning abilities across various knowledge domains and scenarios.

Contribution

The paper presents a novel dataset with complementary sentence pairs, a pairwise accuracy metric, and an adversarial setup to better assess and understand AI commonsense reasoning.

Findings

Baseline models perform significantly below human accuracy.

The dataset reveals gaps in current AI commonsense reasoning capabilities.

Complementary sentence pairs challenge models to improve reasoning robustness.

Abstract

Commonsense reasoning is intuitive for humans but has been a long-term challenge for artificial intelligence (AI). Recent advancements in pretrained language models have shown promising results on several commonsense benchmark datasets. However, the reliability and comprehensiveness of these benchmarks towards assessing model's commonsense reasoning ability remains unclear. To this end, we introduce a new commonsense reasoning benchmark dataset comprising natural language true/false statements, with each sample paired with its complementary counterpart, resulting in 4k sentence pairs. We propose a pairwise accuracy metric to reliably measure an agent's ability to perform commonsense reasoning over a given situation. The dataset is crowdsourced and enhanced with an adversarial model-in-the-loop setup to incentivize challenging samples. To facilitate a systematic analysis of commonsense capabilities, we design our dataset along the dimensions of knowledge domains, reasoning scenarios and numeracy. Experimental results demonstrate that our strongest baseline (UnifiedQA-3B), after fine-tuning, achieves ~71% standard accuracy and ~51% pairwise accuracy, well below human performance (~95% for both metrics). The dataset is available at https://github.com/PlusLabNLP/Com2Sense.