Lune

ACL Findings2022

What to Learn, and How: Toward Effective Learning from Rationales

Samuel Carton, Surya Kanoria, Chenhao Tan

Abstract

Learning from rationales seeks to augment model prediction accuracy using humanannotated rationales (i.e. subsets of input tokens) that justify their chosen labels, often in the form of intermediate or multitask supervision. While intuitive, this idea has proven elusive in practice. We make two observations about human rationales via empirical analyses: 1) maximizing rationale supervision accuracy is not necessarily the optimal objective for improving model accuracy; 2) human rationales vary in whether they provide sufficient information for the model to exploit for prediction. Building on these insights, we propose several novel loss functions and learning strategies, and evaluate their effectiveness on three datasets with human rationales. Our results demonstrate consistent improvements over baselines in both label and rationale accuracy, including a 3% accuracy improvement on MultiRC. Our work highlights the importance of understanding properties of human explanations and exploiting them accordingly in model training.

Ask Lune

Know this paper in minutes, not an afternoon.

Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.

5 top-tier papers have cited it since. Know what they found before your reviewers do.

Questions to start from

Lune can draw on

the full paper · the top-tier papers that cite it · related work

Get the answer

Free to start. No credit card required.

Lineage

  • Neglected Free Lunch - Learning Image Classifiers Using Annotation Byproducts cites Robustness to Spurious Correlations via Human Annotations
  • Learn from A Rationalist: Distilling Intermediate Interpretable Rationales cites Language Models are Few-Shot Learners
  • Learn from A Rationalist: Distilling Intermediate Interpretable Rationales cites Towards Faithful Explanations: Boosting Rationalization with Shortcuts Discovery
  • Regularizing Black-box Models for Improved Interpretability cites Interpretations are Useful: Penalizing Explanations to Align Neural Networks with Prior Knowledge
  • Towards Faithful Explanations: Boosting Rationalization with Shortcuts Discovery cites Invariant Rationalization
Builds onCited byLeft to right by year; the more cited, the higher.

Cited by top-tier papers5

  1. Selective Explanations: Leveraging Human Input to Align Explainable AIVivian Lai, Yiming Zhang, Chacha Chen, Q. Vera Liao et al.CSCW 2023 · 49 citations
  2. Towards Faithful Explanations: Boosting Rationalization with Shortcuts DiscoveryLinan Yue, Qi Liu, Yichao Du, Li Wang et al.ICLR 2024 · 10 citations
  3. Logical Reasoning with Span-Level Predictions for Interpretable and Robust NLI ModelsJoe Stacey, Pasquale Minervini, Haim Dubossarsky, Marek ReiEMNLP 2022 · 5 citations
  4. Neglected Free Lunch - Learning Image Classifiers Using Annotation ByproductsDongyoon Han, Junsuk Choe, Seonghyeok Chun, John Joon Young Chung et al.ICCV 2023 · 4 citations
  5. Learn from A Rationalist: Distilling Intermediate Interpretable RationalesJiayi Dai, Randy GoebelICML 2026

Builds on14

  1. Language Models are Few-Shot LearnersTom B. Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah et al.NeurIPS 2020 · 64,255 citations
  2. Interpretations are Useful: Penalizing Explanations to Align Neural Networks with Prior KnowledgeLaura Rieger, Chandan Singh, W. James Murdoch, Bin YuICML 2020 · 249 citations
  3. Invariant RationalizationShiyu Chang, Yang Zhang, Mo Yu, Tommi S. JaakkolaICML 2020 · 232 citations
  4. Robustness to Spurious Correlations via Human AnnotationsMegha Srivastava, Tatsunori B. Hashimoto, Percy LiangICML 2020 · 103 citations
  5. Regularizing Black-box Models for Improved InterpretabilityGregory Plumb, Maruan Al-Shedivat, Ángel Alexander Cabrera, Adam Perer et al.NeurIPS 2020 · 90 citations

Related papers

  1. Few Shot Rationale Generation using Self-Training with Dual TeachersAditya Srikanth Veerubhotla, Lahari Poddar, Jun Yin, György Szarvas et al.ACL Findings 2023 · 1 citation
  2. Hop, Union, Generate: Explainable Multi-hop Reasoning without Rationale SupervisionWenting Zhao, Justin T. Chiu, Claire Cardie, Alexander M. RushEMNLP 2023 · 4 citations
  3. Zero-Shot Rationalization by Multi-Task Transfer Learning from Question AnsweringPo-Nien Kung, Tse-Hsuan Yang, Yi-Cheng Chen, Sheng-Siang Yin et al.EMNLP Findings 2020 · 4 citations
  4. LIREx: Augmenting Language Inference with Relevant ExplanationsXinyan Zhao, V. G. Vinod VydiswaranAAAI 2021 · 41 citations
  5. MARTA: Leveraging Human Rationales for Explainable Text ClassificationInes Arous, Ljiljana Dolamic, Jie Yang, Akansha Bhardwaj et al.AAAI 2021 · 47 citations
Dusk over the sea between two cliffs drawn in fine vertical lines