Doing Good or Doing Right? Exploring the Weakness of Commonsense Causal Reasoning Models

TL;DR

This paper examines the limitations of current pretrained language models in causal reasoning tasks, revealing their vulnerability to semantic bias and proposing a regularization method to improve robustness and generalization.

Contribution

It identifies semantic similarity bias as a key weakness in COPA models and introduces a simple regularization technique to enhance causal reasoning capabilities.

Findings

Regularization improves model robustness on biased datasets

Models generalize better with the proposed method

Enhanced performance on the challenging BCOPA-CE dataset

Abstract

Pretrained language models (PLM) achieve surprising performance on the Choice of Plausible Alternatives (COPA) task. However, whether PLMs have truly acquired the ability of causal reasoning remains a question. In this paper, we investigate the problem of semantic similarity bias and reveal the vulnerability of current COPA models by certain attacks. Previous solutions that tackle the superficial cues of unbalanced token distribution still encounter the same problem of semantic bias, even more seriously due to the utilization of more training data. We mitigate this problem by simply adding a regularization loss and experimental results show that this solution not only improves the model's generalization ability, but also assists the models to perform more robustly on a challenging dataset, BCOPA-CE, which has unbiased token distribution and is more difficult for models to distinguish cause and effect.