e-SNLI: Natural Language Inference with Natural Language Explanations
Oana-Maria Camburu1 Tim Rocktäschel2 Thomas Lukasiewicz1,3 Phil Blunsom1,4 {oana-maria.camburu, thomas.lukasiewicz, 1Department of Computer Science University of Oxford 2Department of Computer Science University College London 3Alan Turing Institute London UK 4DeepMind London UK
Abstract
In order for machine learning to garner widespread public adoption, models must be able to provide interpretable and robust explanations for their decisions, as well as learn from human-provided explanations at train time. In this work, we extend the Stanford Natural Language Inference dataset with an additional layer of human-annotated natural language explanations of the entailment relations. We further implement models that incorporate these explanations into their training process and output them at test time. We show how our corpus of explanations, which we call e-SNLI, can be used for various goals, such as obtaining full sentence justifications of a model’s decisions, improving universal sentence representations and transferring to out-of-domain NLI datasets. Our dataset111https://github.com/OanaMariaCamburu/e-SNLI thus opens up a range of research directions for using natural language explanations, both for improving models and for asserting their trust.
中文速览
自然语言推理（NLI）模型通常只给出分类标签，却无法告诉用户"为什么"——这既妨碍可信度，也让模型容易依赖数据集里的虚假统计规律。为此，研究者在广泛使用的SNLI数据集上众包收集了57万条人工撰写的自然语言解释，构建出e-SNLI数据集，并在此基础上训练能同步输出标签与文字理由的神经网络模型。实验证明，模型仅凭假设句（hypothesis）生成正确解释的概率比生成正确标签低约十倍，说明解释比标签更难被虚假相关性"蒙混过关"；同时，把解释信号引入训练还能提升通用句子表示（universal sentence representation）的质量，并改善对领域外NLI数据集的迁移效果。e-SNLI是首个大规模自由文本解释语料库，为研究"让模型既能做决策又能讲清楚原因"提供了重要的数据基础和方法参考。
原文 arXiv:1812.01193;中英对照 + 大白话阅读 https://aha.fim.ai/paper/1812.01193v2