# Natural Context Drift Undermines the Natural Language Understanding of Large Language Models

arXiv: https://arxiv.org/abs/2509.01093 · HTML: https://mlsanity.com/paper/2509.01093/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 22 of its 39 references that are in the map.
- First posted on arXiv 1 Sep 2025.

## Builds on (most cited first, 22 of 22)

- [Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks](https://mlsanity.com/paper/1908.10084/) (2019)
- [Efficient Memory Management for Large Language Model Serving with PagedAttention](https://mlsanity.com/paper/2309.06180/) (2023)
- [SQuAD: 100,000+ Questions for Machine Comprehension of Text](https://mlsanity.com/paper/1606.05250/) (2016)
- [HotpotQA: A Dataset for Diverse, Explainable Multi-hop Question Answering](https://mlsanity.com/paper/1809.09600/) (2018)
- [The Pile: An 800GB Dataset of Diverse Text for Language Modeling](https://mlsanity.com/paper/2101.00027/) (2020)
- [BoolQ: Exploring the Surprising Difficulty of Natural Yes/No Questions](https://mlsanity.com/paper/1905.10044/) (2019)
- [Know What You Don't Know: Unanswerable Questions for SQuAD](https://mlsanity.com/paper/1806.03822/) (2018)
- [TinyLlama: An Open-Source Small Language Model](https://mlsanity.com/paper/2401.02385/) (2024)
- [OLMo: Accelerating the Science of Language Models](https://mlsanity.com/paper/2402.00838/) (2024)
- [C-Pack: Packed Resources For General Chinese Embeddings](https://mlsanity.com/paper/2309.07597/) (2023)
- [Dolma: an Open Corpus of Three Trillion Tokens for Language Model Pretraining Research](https://mlsanity.com/paper/2402.00159/) (2024)
- [Sentence-T5: Scalable Sentence Encoders from Pre-trained Text-to-Text Models](https://mlsanity.com/paper/2108.08877/) (2021)
- [Infini-gram: Scaling Unbounded n-gram Language Models to a Trillion Tokens](https://mlsanity.com/paper/2401.17377/) (2024)
- [Beat the AI: Investigating Adversarial Human Annotation for Reading Comprehension](https://mlsanity.com/paper/2002.00293/) (2020)
- [Measure and Improve Robustness in NLP Models: A Survey](https://mlsanity.com/paper/2112.08313/) (2021)
- [Modern Question Answering Datasets and Benchmarks: A Survey](https://mlsanity.com/paper/2206.15030/) (2022)
- [WikiWhy: Answering and Explaining Cause-and-Effect Questions](https://mlsanity.com/paper/2210.12152/) (2022)
- [A Framework for Evaluation of Machine Reading Comprehension Gold Standards](https://mlsanity.com/paper/2003.04642/) (2020)
- [Guiding LLM to Fool Itself: Automatically Manipulating Machine Reading Comprehension Shortcut Triggers](https://mlsanity.com/paper/2310.18360/) (2023)
- [Seemingly Plausible Distractors in Multi-Hop Reasoning: Are Large Language Models Attentive Readers?](https://mlsanity.com/paper/2409.05197/) (2024)
- [NeoQA: Evidence-based Question Answering with Generated News Events](https://mlsanity.com/paper/2505.05949/) (2025)
- [Pay Attention to Real World Perturbations! Natural Robustness Evaluation in Machine Reading Comprehension](https://mlsanity.com/paper/2502.16523/) (2025)

## Built on by (most cited first, 1 of 1)

- [Embodied Task Planning via Graph-Informed Action Generation with Large Language Models](https://mlsanity.com/paper/2601.21841/) (2026)

## Abstract (arXiv)

How does the natural evolution of context paragraphs affect question answering in generative Large Language Models (LLMs)? To investigate this, we propose a framework for curating naturally evolved, human-edited variants of reading passages from contemporary QA benchmarks and for analyzing LLM performance across a range of semantic similarity scores, which quantify how closely each variant aligns with content seen during pretraining. Using this framework, we evaluate six QA datasets and eight LLMs with publicly available training data. Our experiments reveal that LLM performance declines as reading passages naturally diverge from the versions encountered during pretraining-even when the question and all necessary information remains present at inference time. For instance, average model accuracy on BoolQ drops by over 30% from the highest to lowest similarity bins, with slopes exceeding 70 across several LLMs. These findings suggest that natural text evolution poses a significant challenge to the language understanding capabilities of LLMs.