Learning the Difference that Makes a Difference with Counterfactually-Augmented Data

Abstract

Despite alarm over the reliance of machine learning systems on so-called spurious patterns, the term lacks coherent meaning in standard statistical frameworks. However, the language of causality offers clarity: spurious associations are due to confounding (e.g., a common cause), but not direct or indirect causal effects. In this paper, we focus on natural language processing, introducing methods and resources for training models less sensitive to spurious patterns. Given documents and their initial labels, we task humans with revising each document so that it (i) accords with a counterfactual target label; (ii) retains internal coherence; and (iii) avoids unnecessary changes. Interestingly, on sentiment analysis and natural language inference tasks, classifiers trained on original data fail on their counterfactually-revised counterparts and vice versa. Classifiers trained on combined datasets perform remarkably well, just shy of those specialized to either domain. While classifiers trained on either original or manipulated data alone are sensitive to spurious features (e.g., mentions of genre), models trained on the combined data are less sensitive to this signal. Both datasets are publicly available.

arXiv commentPublished at ICLR 2020

References in corpus (6)

• Scikit-learn: Machine Learning in Python
• Teaching Machines to Read and Comprehend
• Measuring the tendency of CNNs to Learn Surface Statistical Regularities
• Gender Bias in Coreference Resolution: Evaluation and Debiasing Methods
• Gender Bias in Neural Natural Language Processing
• What do Deep Networks Like to Read?

Cited by in corpus (16)

• Intelligent Computing: The Latest Advances, Challenges and Future
• The Hateful Memes Challenge: Detecting Hate Speech in Multimodal Memes
• Adversarial NLI: A New Benchmark for Natural Language Understanding
• Detecting Hate Speech in Memes Using Multimodal Deep Learning Approaches: Prize-winning solution to Hateful Memes Challenge
• The Effect of Natural Distribution Shift on Question Answering Models
• A Unifying Causal Framework for Analyzing Dataset Shift-stable Learning Algorithms
• CommonsenseVIS: Visualizing and Understanding Commonsense Reasoning Capabilities of Natural Language Models
• Putting Humans in the Natural Language Processing Loop: A Survey
• Causal Inference in Natural Language Processing: Estimation, Prediction, Interpretation and Beyond
• Unsupervised Editing for Counterfactual Stories
• A Survey on Out-of-Distribution Evaluation of Neural NLP Models
• SparCAssist: A Model Risk Assessment Assistant Based on Sparse Generated Counterfactuals
• A Data-centric Framework for Improving Domain-specific Machine Reading Comprehension Datasets
• Geometry matters: Exploring language examples at the decision boundary
• GECOBench: A Gender-Controlled Text Dataset and Benchmark for Quantifying Biases in Explanations
• Adversarial Examples Generation for Reducing Implicit Gender Bias in Pre-trained Models