When Choosing Plausible Alternatives, Clever Hans can be Clever

TL;DR

This paper investigates whether BERT and RoBERTa's high performance on COPA is due to superficial cues and introduces Balanced COPA to evaluate models without these cues, revealing different reliance patterns.

Contribution

The paper identifies superficial cues in COPA, introduces Balanced COPA to mitigate these cues, and analyzes model reliance, highlighting differences between BERT and RoBERTa.

Findings

BERT exploits superficial cues in COPA but learns the task when cues are removed.

RoBERTa does not rely on superficial cues in COPA.

Balanced COPA reduces superficial cue exploitation, providing a fairer evaluation.

Abstract

Pretrained language models, such as BERT and RoBERTa, have shown large improvements in the commonsense reasoning benchmark COPA. However, recent work found that many improvements in benchmarks of natural language understanding are not due to models learning the task, but due to their increasing ability to exploit superficial cues, such as tokens that occur more often in the correct answer than the wrong one. Are BERT's and RoBERTa's good performance on COPA also caused by this? We find superficial cues in COPA, as well as evidence that BERT exploits these cues. To remedy this problem, we introduce Balanced COPA, an extension of COPA that does not suffer from easy-to-exploit single token cues. We analyze BERT's and RoBERTa's performance on original and Balanced COPA, finding that BERT relies on superficial cues when they are present, but still achieves comparable performance once they are made ineffective, suggesting that BERT learns the task to a certain degree when forced to. In contrast, RoBERTa does not appear to rely on superficial cues.