# Spontaneous Context Restoration: How Language Models Recover from Corrupted Inputs

arXiv: https://arxiv.org/abs/2609.35475 · HTML: https://mlsanity.com/paper/2609.35475/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Placed in Knowledge and representations inside LLMs, one of 13 directions in Mechanistic interpretability.
- Builds on 27 of its 37 references that are in the map.
- First posted on arXiv 28 Sep 2026.

## Placed in

- Mechanistic interpretability / [Knowledge and representations inside LLMs](https://mlsanity.com/mechinterp/knowledge-and-representations-inside-llms/)

## Builds on (most cited first, 27 of 27)

- [Think you have Solved Question Answering? Try ARC, the AI2 Reasoning Challenge](https://mlsanity.com/paper/1803.05457/) (2018)
- [SQuAD: 100,000+ Questions for Machine Comprehension of Text](https://mlsanity.com/paper/1606.05250/) (2016)
- [Lost in the Middle: How Language Models Use Long Contexts](https://mlsanity.com/paper/2307.03172/) (2023)
- [Locating and Editing Factual Associations in GPT](https://mlsanity.com/paper/2202.05262/) (2022)
- [Rethinking the Role of Demonstrations: What Makes In-Context Learning Work?](https://mlsanity.com/paper/2202.12837/) (2022)
- [Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 small](https://mlsanity.com/paper/2211.00593/) (2022)
- [Right for the Wrong Reasons: Diagnosing Syntactic Heuristics in Natural Language Inference](https://mlsanity.com/paper/1902.01007/) (2019)
- [The Geometry of Truth: Emergent Linear Structure in Large Language Model Representations of True/False Datasets](https://mlsanity.com/paper/2310.06824/) (2023)
- [Towards Best Practices of Activation Patching in Language Models: Metrics and Methods](https://mlsanity.com/paper/2309.16042/) (2023)
- [How to use and interpret activation patching](https://mlsanity.com/paper/2404.15255/) (2024)
- [Not All Language Model Features Are One-Dimensionally Linear](https://mlsanity.com/paper/2405.14860/) (2024)
- [Localizing Model Behavior with Path Patching](https://mlsanity.com/paper/2304.05969/) (2023)
- [How Context Affects Language Models' Factual Predictions](https://mlsanity.com/paper/2005.04611/) (2020)
- [The Hydra Effect: Emergent Self-repair in Language Model Computations](https://mlsanity.com/paper/2307.15771/) (2023)
- [Robust Prompt Optimization for Defending Language Models Against Jailbreaking Attacks](https://mlsanity.com/paper/2401.17263/) (2024)
- [The Remarkable Robustness of LLMs: Stages of Inference?](https://mlsanity.com/paper/2406.19384/) (2024)
- [Measure and Improve Robustness in NLP Models: A Survey](https://mlsanity.com/paper/2112.08313/) (2021)
- [DisentQA: Disentangling Parametric and Contextual Knowledge with Counterfactual Question Answering](https://mlsanity.com/paper/2211.05655/) (2022)
- [Transformers represent belief state geometry in their residual stream](https://mlsanity.com/paper/2405.15943/) (2024)
- [Interpreting and Improving Large Language Models in Arithmetic Calculation](https://mlsanity.com/paper/2409.01659/) (2024)
- [Can Language Models Perform Robust Reasoning in Chain-of-thought Prompting with Noisy Rationales?](https://mlsanity.com/paper/2410.23856/) (2024)
- [Competition of Mechanisms: Tracing How Language Models Handle Facts and Counterfactuals](https://mlsanity.com/paper/2402.11655/) (2024)
- [Explorations of Self-Repair in Language Models](https://mlsanity.com/paper/2402.15390/) (2024)
- [Interpreting Arithmetic Mechanism in Large Language Models through Comparative Neuron Analysis](https://mlsanity.com/paper/2409.14144/) (2024)
- [Robustness of Large Language Models to Perturbations in Text](https://mlsanity.com/paper/2407.08989/) (2024)
- [Evaluating Concurrent Robustness of Language Models Across Diverse Challenge Sets](https://mlsanity.com/paper/2311.08662/) (2023)
- [Improving the Robustness of Large Language Models for Code Tasks via Fine-tuning with Perturbed Data](https://mlsanity.com/paper/2602.11411/) (2026)

## Abstract (arXiv)

Language models sometimes produce correct outputs even when their inputs are corrupted by deletion, replacement, or misspelling. We study the internal processes accompanying this behavior, which we call context restoration, in controlled attention-only transformers and five pretrained LLMs (1B-32B parameters) across arithmetic, reading comprehension, and multiple-choice reasoning tasks. In the attention-only transformers, restoration emerges spontaneously despite training exclusively on clean sequences, without corruption training or an explicit denoising objective. We find that context restoration follows a two-phase process: early layers localize effects associated with repair at corrupted positions, while later layers accumulate these effects at uncorrupted positions through the residual stream and ultimately concentrate them at the output position. Repair outcome is predictable from hidden states: cosine alignment with the clean state is highly predictive in attention-only models, while linear probes recover additional information in pretrained LLMs. A linear probe using only the corrupted prompt's first-block hidden state predicts failure with mean ROC-AUC 0.78. This enables failure triage under matched or even partially shifted deployment conditions and may reduce unnecessary verification or computation. Failed examples also show substantially greater nonlinearity along corruption directions. Moderate-corruption finetuning increases corruption tolerance while simultaneously reducing displacement-normalized linearization error, associating improved robustness with a more nearly linear response to corruption.