Learning from others' mistakes: Avoiding dataset biases without modeling them
This work is significant for NLP researchers and practitioners as it offers a general method to improve model robustness and out-of-distribution performance without prior knowledge of dataset biases, which is a common and challenging problem.
This paper addresses the problem of NLP models learning dataset biases instead of task-relevant features. The authors propose a method that leverages the errors of limited-capacity models, which primarily exploit biases, to train a more robust model in a product of experts framework, thereby avoiding the need to explicitly identify or model these biases.
State-of-the-art natural language processing (NLP) models often learn to model dataset biases and surface form correlations instead of features that target the intended underlying task. Previous work has demonstrated effective methods to circumvent these issues when knowledge of the bias is available. We consider cases where the bias issues may not be explicitly identified, and show a method for training models that learn to ignore these problematic correlations. Our approach relies on the observation that models with limited capacity primarily learn to exploit biases in the dataset. We can leverage the errors of such limited capacity models to train a more robust model in a product of experts, thus bypassing the need to hand-craft a biased model. We show the effectiveness of this method to retain improvements in out-of-distribution settings even if no particular bias is targeted by the biased model.