NILE : Natural Language Inference with Faithful Natural Language Explanations

By Sawan Kumar, Partha P. Talukdar. 2020

174 citations recorded by the source corpus.

The recent growth in the popularity and success of deep learning models on NLP classification tasks has accompanied the need for generating some form of natural language explanation of the predicted labels. Such generated natural language (NL) explanations are expected to be faithful, i.e., they should correlate well with the model’s internal decision making. In this work, we focus on the task of natural language inference (NLI) and address the following question: can we build NLI systems which produce labels with high accuracy, while also generating faithful explanations of its decisions? We propose Natural-language Inference over Label-specific Explanations (NILE), a novel NLI method which utilizes auto-generated label-specific NL explanations to produce labels along with its faithful explanation. We demonstrate NILE’s effectiveness over previously reported methods through automated and human evaluation of the produced labels and explanations. Our evaluation of NILE also supports the claim that accurate systems capable of providing testable explanations of their decisions can be designed. We discuss the faithfulness of NILE’s explanations in terms of sensitivity of the decisions to the corresponding explanations. We argue that explicit evaluation of faithfulness, in addition to label and explanation accuracy, is an important step in evaluating model’s explanations. Further, we demonstrate that task-specific probes are necessary to establish such sensitivity.

Subfields: Artificial Intelligence, Natural Language Processing

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Natural Language Processing

Citations in the corpus

Few-Shot Out-of-Domain Transfer Learning of Natural Language Explanations (2021, background): “Zero-shot in-domain transfer of NLEs (between datasets of the same task) has been done, e.g., by Camburu et al. (2018); Kumar and Talukdar (2020) 2022) use prompt engineering for few-shot out-of-domain transfer of NLEs for the scarce-label setup.”
When Can Models Learn From Explanations? A Formal Framework for Understanding the Roles of Explanation Data (2021, neutral): “2 (Camburu et al., 2018; Rajani et al., 2019; Kumar & Talukdar, 2020; Hase et al., 2020; Wiegreffe et al., 2020).”
Measuring Association Between Labels and Free-Text Rationales (2020, background): “first need to implicitly predict a label (Kumar and Talukdar, 2020; Jacovi and Goldberg, 2020a).”
Faithful Reasoning Using Large Language Models (2022, background): “Finally, while several works have informally introduced the notion of faithful reasoning (Bostrom et al., 2022; Gupta et al., 2022; Kumar and Talukdar, 2020), we have related this more precisely to the deﬁnition of valid reasoning in logic.”
e-ViL: A Dataset and Benchmark for Natural Language Explanations in Vision-Language Tasks (2021, background): “First adoptions of NLEs have been in image classification [23] and were further extended to self-driving cars [26], visual question-answering [37], and natural language processing [11; 39; 9; 6; 27; 35; 28; 12].”
Interpretable deep learning: interpretation, interpretability, trustworthiness, and beyond (2021, methodology): “A joint output-explanation model is trained to produce an prediction and simultaneously generate an explanation for the reason of that prediction [14, 93, 109].”
Knowledge-Grounded Self-Rationalization via Extractive and Natural Language Explanations (2021, methodology): “We also compare with works that model NLEs speciﬁcally for a dataset: WT5 for ComVE, NILE (Kumar & Talukdar, 2020) for e-SNLI, and CAGE (Rajani et al., 2019) for COSe.”
Natural Language Inference with a Human Touch: Using Human Explanations to Guide Model Attention (2021, methodology): “Human explanations have been found to improve performance on a range of tasks (Rajani et al., 2019; Andreas et al., 2018; Mu et al., 2020; Liang et al., 2020), but this has largely not been the case in NLI (Hase and Bansal, 2021; Kumar and Talukdar, 2020; Camburu et al., 2018).”
Supervising Model Attention with Human Explanations for Robust Natural Language Inference (2021, methodology): “Kumar and Talukdar (2020) generate explanations spe-ciﬁc to each class, using these explanations along with the premise and hypothesis to predict the NLI class.”
Leakage-Adjusted Simulatability: Can Models Generate Non-Trivial Explanations of Their Behavior in Natural Language? (2020, methodology): “Kumar and Talukdar (2020) introduce label-speciﬁc versions of the method in Rajani et al. (2019), one of which shares the graphical structure of our ST-R A model.”
A Study of Automatic Metrics for the Evaluation of Natural Language Explanations (2021, background): “…whether general, domain-independent evaluation metrics within NLG are sensitive enough to capture the peculiarities inherent in NL explanations (Kumar and Talukdar, 2020), such as causality; or whether NL explanations constitute a sui-generis category, thus requiring their own automatic…”
Teach Me to Explain: A Review of Datasets for Explainable NLP (2021, background): “However, it is possible that the model learns to ignore the semantics of the explanation and instead makes predictions on the basis of the explanation’s template type (Kumar and Talukdar, 2020; Jacovi and Goldberg, 2021).”
Using Interactive Feedback to Improve the Accuracy and Explainability of Question Answering Systems Post-Deployment (2022, background): “…collecting and harnessing explanation data for language understanding and reasoning tasks, such as natural language inference (Camburu et al., 2018; Kumar and Talukdar, 2020), commonsense question answering (Rajani et al., 2019), document classiﬁcation (Srivastava et al., 2017), relation clas-522…”
Multi-document Summarization via Deep Learning Techniques: A Survey (2020, background): “The idea of using adversarial examples to examine the robustness of a DNN model originated from research in Computer Vision [144] and was introduced in NLP by Jia et al. [70].”
Rationale-Inspired Natural Language Explanations with Commonsense (2021, methodology): “For e-SNLI and COSe, we use two additional SOTA models: NILE [21] and GPT-2 [36], respectively.”
REV: Information-Theoretic Evaluation of Free-Text Rationales (2022, background): “Compared to extractive rationales, free-text rationales explain model predictions in a more human-like way and fill the gap in explaining reasoning tasks (Camburu et al., 2018; Narang et al., 2020; Rajani et al., 2019; Kumar and Talukdar, 2020; Brahman et al., 2021).”
Learning to Rationalize for Nonmonotonic Reasoning with Distant Supervision (2020, methodology): “…in some cases the selected rationale is not sufﬁcient to make the prediction without the input (Wiegreffe, Sarah and Marasovi´c, Ana, and Smith, Noah A. 2020), while in others, label-speciﬁc rationale templates may make the label prediction trivial given the rationale (Kumar and Talukdar 2020).”
multiPRover: Generating Multiple Proofs for Improved Interpretability in Rule Reasoning (2021, background): “…of the input text explain a prediction, (2) free-form or natural language explanations (Camburu et al., 2018; Ra-jani et al., 2019; Zhang et al., 2020; Kumar and Talukdar, 2020) that are not constrained to the input, and (3) structured explanations that range from semi-structured text (Ye et al.,…”
Natural Language Rationales with Full-Stack Visual Reasoning: From Pixels to Semantic Frames to Commonsense Graphs (2020, methodology): “Kumar and Talukdar (2020) highlight that this approach resembles post-hoc methods with the label and rationale being produced jointly (the end-toend predict-then-explain setting).”