# Refining Language Models with Compositional Explanations

arXiv: https://arxiv.org/abs/2103.10415 · HTML: https://mlsanity.com/paper/2103.10415/

## Key facts

- Cited by 27 papers in the map (data through 14 Oct 2026).
- Builds on 31 of its 75 references that are in the map.
- First posted on arXiv 18 Mar 2021.

## Builds on (most cited first, 30 of 31)

- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [Adam: A Method for Stochastic Optimization](https://mlsanity.com/paper/1412.6980/) (2014)
- [RoBERTa: A Robustly Optimized BERT Pretraining Approach](https://mlsanity.com/paper/1907.11692/) (2019)
- [Distilling the Knowledge in a Neural Network](https://mlsanity.com/paper/1503.02531/) (2015)
- [Axiomatic Attribution for Deep Networks](https://mlsanity.com/paper/1703.01365/) (2017)
- [ALBERT: A Lite BERT for Self-supervised Learning of Language Representations](https://mlsanity.com/paper/1909.11942/) (2019)
- [Unsupervised Domain Adaptation by Backpropagation](https://mlsanity.com/paper/1409.7495/) (2014)
- [Moment Matching for Multi-Source Domain Adaptation](https://mlsanity.com/paper/1812.01754/) (2018)
- [A Survey Of Methods For Explaining Black Box Models](https://mlsanity.com/paper/1802.01933/) (2018)
- [Do We Really Need to Access the Source Data? Source Hypothesis Transfer for Unsupervised Domain Adaptation](https://mlsanity.com/paper/2002.08546/) (2020)
- [Ups and Downs: Modeling the Visual Evolution of Fashion Trends with One-Class Collaborative Filtering](https://mlsanity.com/paper/1602.01585/) (2016)
- [DeCAF: A Deep Convolutional Activation Feature for Generic Visual Recognition](https://mlsanity.com/paper/1310.1531/) (2013)
- [Return of Frustratingly Easy Domain Adaptation](https://mlsanity.com/paper/1511.05547/) (2015)
- [Right for the Right Reasons: Training Differentiable Models by Constraining their Explanations](https://mlsanity.com/paper/1703.03717/) (2017)
- [Explicit Inductive Bias for Transfer Learning with Convolutional Networks](https://mlsanity.com/paper/1802.01483/) (2018)
- [Hate Speech Dataset from a White Supremacy Forum](https://mlsanity.com/paper/1809.04444/) (2018)
- [Central Moment Discrepancy (CMD) for Domain-Invariant Representation Learning](https://mlsanity.com/paper/1702.08811/) (2017)
- [Visualizing Deep Neural Network Decisions: Prediction Difference Analysis](https://mlsanity.com/paper/1702.04595/) (2017)
- [Interpretations are useful: penalizing explanations to align neural networks with prior knowledge](https://mlsanity.com/paper/1909.13584/) (2019)
- [Using millions of emoji occurrences to learn any-domain representations for detecting sentiment, emotion and sarcasm](https://mlsanity.com/paper/1708.00524/) (2017)
- [Contextualizing Hate Speech Classifiers with Post-hoc Explanation](https://mlsanity.com/paper/2005.02439/) (2020)
- [Incorporating Priors with Feature Attribution on Text Classification](https://mlsanity.com/paper/1906.08286/) (2019)
- [Training Classifiers with Natural Language Explanations](https://mlsanity.com/paper/1805.03818/) (2018)
- [FIND: Human-in-the-Loop Debugging Deep Text Classifiers](https://mlsanity.com/paper/2010.04987/) (2020)
- [Transfer Learning for Speech and Language Processing](https://mlsanity.com/paper/1511.06066/) (2015)
- [Regularizing Black-box Models for Improved Interpretability](https://mlsanity.com/paper/1902.06787/) (2019)
- [Improving performance of deep learning models with axiomatic attribution priors and expected gradients](https://mlsanity.com/paper/1906.10670/) (2019)
- [Neural Domain Adaptation for Biomedical Question Answering](https://mlsanity.com/paper/1706.03610/) (2017)
- [Teaching Machine Comprehension with Compositional Explanations](https://mlsanity.com/paper/2005.00806/) (2020)
- [Learning from Explanations with Neural Execution Tree](https://mlsanity.com/paper/1911.01352/) (2019)

## Built on by (most cited first, 27 of 27)

- [Chain-of-Thought Prompting Elicits Reasoning in Large Language Models](https://mlsanity.com/paper/2201.11903/) (2022)
- [The Unreliability of Explanations in Few-shot Prompting for Textual Reasoning](https://mlsanity.com/paper/2205.03401/) (2022)
- [Cumulative Reasoning with Large Language Models](https://mlsanity.com/paper/2308.04371/) (2023)
- [Explanation-Based Human Debugging of NLP Models: A Survey](https://mlsanity.com/paper/2104.15135/) (2021)
- [Going Beyond XAI: A Systematic Survey for Explanation-Guided Learning](https://mlsanity.com/paper/2212.03954/) (2022)
- [Sociodemographic Bias in Language Models: A Survey and Forward Path](https://mlsanity.com/paper/2306.08158/) (2023)
- [MCC-KD: Multi-CoT Consistent Knowledge Distillation](https://mlsanity.com/paper/2310.14747/) (2023)
- [On Behalf of the Stakeholders: Trends in NLP Model Interpretability in the Era of LLMs](https://mlsanity.com/paper/2407.19200/) (2024)
- [Multi-Defendant Legal Judgment Prediction via Hierarchical Reasoning](https://mlsanity.com/paper/2312.05762/) (2023)
- [Meta Prompting for AI Systems](https://mlsanity.com/paper/2311.11482/) (2023)
- [A survey on improving NLP models with human explanations](https://mlsanity.com/paper/2204.08892/) (2022)
- [Saliency Map Verbalization: Comparing Feature Importance Representations from Model-free and Instruction-based Methods](https://mlsanity.com/paper/2210.07222/) (2022)
- [Knowledge-augmented Deep Learning and Its Applications: A Survey](https://mlsanity.com/paper/2212.00017/) (2022)
- [What Will My Model Forget? Forecasting Forgotten Examples in Language Model Refinement](https://mlsanity.com/paper/2402.01865/) (2024)
- [XMD: An End-to-End Framework for Interactive Explanation-Based Debugging of NLP Models](https://mlsanity.com/paper/2210.16978/) (2022)
- [Informing clinical assessment by contextualizing post-hoc explanations of risk prediction models in type-2 diabetes](https://mlsanity.com/paper/2302.05752/) (2023)
- [IFAN: An Explainability-Focused Interaction Framework for Humans and NLP Models](https://mlsanity.com/paper/2303.03124/) (2023)
- [Clarify: Improving Model Robustness With Natural Language Corrections](https://mlsanity.com/paper/2402.03715/) (2024)
- [Mediators: Conversational Agents Explaining NLP Model Behavior](https://mlsanity.com/paper/2206.06029/) (2022)
- [Assessing Out-of-Domain Language Model Performance from Few Examples](https://mlsanity.com/paper/2210.06725/) (2022)
- [Power of Explanations: Towards automatic debiasing in hate speech detection](https://mlsanity.com/paper/2209.09975/) (2022)
- [Manifestations of Xenophobia in AI Systems](https://mlsanity.com/paper/2212.07877/) (2022)
- [Neglected Free Lunch -- Learning Image Classifiers Using Annotation Byproducts](https://mlsanity.com/paper/2303.17595/) (2023)
- [Dynamically Refined Regularization for Improving Cross-corpora Hate Speech Detection](https://mlsanity.com/paper/2203.12536/) (2022)
- [A Survey on Knowledge-Enhanced Pre-trained Language Models](https://mlsanity.com/paper/2212.13428/) (2022)
- [A look under the hood of the Interactive Deep Learning Enterprise (No-IDLE)](https://mlsanity.com/paper/2406.19054/) (2024)
- [Domain Classification-based Source-specific Term Penalization for Domain Adaptation in Hate-speech Detection](https://mlsanity.com/paper/2209.08681/) (2022)

## Abstract (arXiv)

Pre-trained language models have been successful on text classification tasks, but are prone to learning spurious correlations from biased datasets, and are thus vulnerable when making inferences in a new domain. Prior work reveals such spurious patterns via post-hoc explanation algorithms which compute the importance of input features. Further, the model is regularized to align the importance scores with human knowledge, so that the unintended model behaviors are eliminated. However, such a regularization technique lacks flexibility and coverage, since only importance scores towards a pre-defined list of features are adjusted, while more complex human knowledge such as feature interaction and pattern generalization can hardly be incorporated. In this work, we propose to refine a learned language model for a target domain by collecting human-provided compositional explanations regarding observed biases. By parsing these explanations into executable logic rules, the human-specified refinement advice from a small set of explanations can be generalized to more training examples. We additionally introduce a regularization term allowing adjustments for both importance and interaction of features to better rectify model behavior. We demonstrate the effectiveness of the proposed approach on two text classification tasks by showing improved performance in target domain as well as improved model fairness after refinement.