# Is the MMI Criterion Necessary for Interpretability? Degenerating Non-causal Features to Plain Noise for Self-Rationalization

arXiv: https://arxiv.org/abs/2410.06003 · HTML: https://mlsanity.com/paper/2410.06003/

## Key facts

- Cited by 11 papers in the map (data through 14 Oct 2026).
- Builds on 24 of its 42 references that are in the map.
- First posted on arXiv 8 Oct 2024.

## Builds on (most cited first, 24 of 24)

- [Invariant Risk Minimization](https://mlsanity.com/paper/1907.02893/) (2019)
- [Sanity Checks for Saliency Maps](https://mlsanity.com/paper/1810.03292/) (2018)
- [Towards Faithfully Interpretable NLP Systems: How should we define and evaluate faithfulness?](https://mlsanity.com/paper/2004.03685/) (2020)
- [LESS: Selecting Influential Data for Targeted Instruction Tuning](https://mlsanity.com/paper/2402.04333/) (2024)
- [ERASER: A Benchmark to Evaluate Rationalized NLP Models](https://mlsanity.com/paper/1911.03429/) (2019)
- [Parameterized Explainer for Graph Neural Network](https://mlsanity.com/paper/2011.04573/) (2020)
- [Rationalizing Neural Predictions](https://mlsanity.com/paper/1606.04155/) (2016)
- [The Risks of Invariant Risk Minimization](https://mlsanity.com/paper/2010.05761/) (2020)
- [Evaluating Large Language Models: A Comprehensive Survey](https://mlsanity.com/paper/2310.19736/) (2023)
- [Invariant Rationalization](https://mlsanity.com/paper/2003.09772/) (2020)
- [Rethinking Cooperative Rationalization: Introspective Extraction and Complement Control](https://mlsanity.com/paper/1910.13294/) (2019)
- [ZIN: When and How to Learn Invariance Without Environment Partition?](https://mlsanity.com/paper/2203.05818/) (2022)
- [Understanding Interlocking Dynamics of Cooperative Rationalization](https://mlsanity.com/paper/2110.13880/) (2021)
- [UNIREX: A Unified Learning Framework for Language Model Rationale Extraction](https://mlsanity.com/paper/2112.08802/) (2021)
- [Joint Learning of Label and Environment Causal Independence for Graph Out-of-Distribution Generalization](https://mlsanity.com/paper/2306.01103/) (2023)
- [Learning from the Best: Rationalizing Prediction by Adversarial Information Calibration](https://mlsanity.com/paper/2012.08884/) (2020)
- [D-Separation for Causal Self-Explanation](https://mlsanity.com/paper/2309.13391/) (2023)
- [Towards Trustworthy Explanation: On Causal Rationalization](https://mlsanity.com/paper/2306.14115/) (2023)
- [FR: Folded Rationalization with a Unified Encoder](https://mlsanity.com/paper/2209.08285/) (2022)
- [Enhancing the Rationale-Input Alignment for Self-explaining Rationalization](https://mlsanity.com/paper/2312.04103/) (2023)
- [Decoupled Rationalization with Asymmetric Learning Rates: A Flexible Lipschitz Restraint](https://mlsanity.com/paper/2305.13599/) (2023)
- [Are Shortest Rationales the Best Explanations for Human Understanding?](https://mlsanity.com/paper/2203.08788/) (2022)
- [Towards Faithful Explanations: Boosting Rationalization with Shortcuts Discovery](https://mlsanity.com/paper/2403.07955/) (2024)
- [MARE: Multi-Aspect Rationale Extractor on Unsupervised Rationale Extraction](https://mlsanity.com/paper/2410.03531/) (2024)

## Built on by (most cited first, 11 of 11)

- [When Backdoors Speak: Understanding LLM Backdoor Attacks Through Model-Generated Explanations](https://mlsanity.com/paper/2411.12701/) (2024)
- [Beyond Zero Initialization: Investigating the Impact of Non-Zero Initialization on LoRA Fine-Tuning Dynamics](https://mlsanity.com/paper/2505.23194/) (2025)
- [Breaking Free from MMI: A New Frontier in Rationalization by Probing Input Utilization](https://mlsanity.com/paper/2503.06202/) (2025)
- [Adversarial Cooperative Rationalization: The Risk of Spurious Correlations in Even Clean Datasets](https://mlsanity.com/paper/2505.02118/) (2025)
- [The Panaceas for Improving Low-Rank Decomposition in Communication-Efficient Federated Learning](https://mlsanity.com/paper/2505.23176/) (2025)
- [GNN Explanations that do not Explain and How to find Them](https://mlsanity.com/paper/2601.20815/) (2026)
- [Graph-Guided Textual Explanation Generation Framework](https://mlsanity.com/paper/2412.12318/) (2024)
- [Beyond External Monitors: Enhancing Transparency of Large Language Models for Easier Monitoring](https://mlsanity.com/paper/2502.05242/) (2025)
- [TopInG: Topologically Interpretable Graph Learning via Persistent Rationale Filtration](https://mlsanity.com/paper/2510.05102/) (2025)
- [Interlocking-free Selective Rationalization Through Genetic-based Learning](https://mlsanity.com/paper/2412.10312/) (2024)
- [Direct Preference Optimization for Adaptive Concept-based Explanations](https://mlsanity.com/paper/2505.15626/) (2025)

## Abstract (arXiv)

An important line of research in the field of explainability is to extract a small subset of crucial rationales from the full input. The most widely used criterion for rationale extraction is the maximum mutual information (MMI) criterion. However, in certain datasets, there are spurious features non-causally correlated with the label and also get high mutual information, complicating the loss landscape of MMI. Although some penalty-based methods have been developed to penalize the spurious features (e.g., invariance penalty, intervention penalty, etc) to help MMI work better, these are merely remedial measures. In the optimization objectives of these methods, spurious features are still distinguished from plain noise, which hinders the discovery of causal rationales. This paper aims to develop a new criterion that treats spurious features as plain noise, allowing the model to work on datasets rich in spurious features as if it were working on clean datasets, thereby making rationale extraction easier. We theoretically observe that removing either plain noise or spurious features from the input does not alter the conditional distribution of the remaining components relative to the task label. However, significant changes in the conditional distribution occur only when causal features are eliminated. Based on this discovery, the paper proposes a criterion for Maximizing the Remaining Discrepancy (MRD). Experiments on six widely used datasets show that our MRD criterion improves rationale quality (measured by the overlap with human-annotated rationales) by up to 10.4% as compared to several recent competitive MMI variants. Code: urlhttps://github.com/jugechengzi/Rationalization-MRD.