论文标题
反事实的变量控制，可用于鲁棒和可解释的问题答案
Counterfactual Variable Control for Robust and Interpretable Question Answering
论文作者
论文摘要
在许多情况下，基于神经网络的深度网络答案（QA）模型既不强大，也不是可解释的。例如，在没有任何问题输入的情况下测试的多项选择质量检查模型令人惊讶地“能够”预测大多数正确的选项。在本文中，我们使用因果推断检查了质量检查模型的这种虚假“能力”。我们发现关键是捷径相关性，例如，段落和模型所学的选项之间的单词对齐方式是不可行的。我们提出了一种称为反事实变量控制（CVC）的新方法，该方法明确减轻了任何快捷方式相关性，并保留了稳健质量检查的全面推理。具体而言，我们利用多分支结构，使我们能够在QA的培训过程中解散强大和快捷方式的相关性。然后，我们进行了两种新型的CVC推理方法（在训练有素的模型上），以捕获全面推理作为最终预测的影响。为了进行评估，我们在多选择和跨度QA基准上使用两个BERT主链进行了广泛的实验。结果表明，我们的CVC在质量检查中对各种对抗性攻击具有很高的鲁棒性，同时保持良好的解释能力。
Deep neural network based question answering (QA) models are neither robust nor explainable in many cases. For example, a multiple-choice QA model, tested without any input of question, is surprisingly "capable" to predict the most of correct options. In this paper, we inspect such spurious "capability" of QA models using causal inference. We find the crux is the shortcut correlation, e.g., unrobust word alignment between passage and options learned by the models. We propose a novel approach called Counterfactual Variable Control (CVC) that explicitly mitigates any shortcut correlation and preserves the comprehensive reasoning for robust QA. Specifically, we leverage multi-branch architecture that allows us to disentangle robust and shortcut correlations in the training process of QA. We then conduct two novel CVC inference methods (on trained models) to capture the effect of comprehensive reasoning as the final prediction. For evaluation, we conduct extensive experiments using two BERT backbones on both multi-choice and span-extraction QA benchmarks. The results show that our CVC achieves high robustness against a variety of adversarial attacks in QA while maintaining good interpretation ability.
