# When Context Misleads: Surprisal, Energy and Attention Entropy as Metrics of Coherence Illusions in LLMs

arXiv: https://arxiv.org/abs/2606.21203 · HTML: https://mlsanity.com/paper/2606.21203/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 26 of its 53 references that are in the map.
- First posted on arXiv 19 Jun 2026.

## Builds on (most cited first, 26 of 26)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [Interpretability in the Wild: a Circuit for Indirect Object Identification in GPT-2 small](https://mlsanity.com/paper/2211.00593/) (2022)
- [Probing Classifiers: Promises, Shortcomings, and Advances](https://mlsanity.com/paper/2102.12452/) (2021)
- [Large Language Models Can Be Easily Distracted by Irrelevant Context](https://mlsanity.com/paper/2302.00093/) (2023)
- [Hopfield Networks is All You Need](https://mlsanity.com/paper/2008.02217/) (2020)
- [Making Retrieval-Augmented Language Models Robust to Irrelevant Context](https://mlsanity.com/paper/2310.01558/) (2023)
- [Assessing the Ability of LSTMs to Learn Syntax-Sensitive Dependencies](https://mlsanity.com/paper/1611.01368/) (2016)
- [Dense Associative Memory for Pattern Recognition](https://mlsanity.com/paper/1606.01164/) (2016)
- [Why Does Surprisal From Larger Transformer-Based Language Models Provide a Poorer Fit to Human Reading Times?](https://mlsanity.com/paper/2212.12131/) (2022)
- [EuroLLM: Multilingual Language Models for Europe](https://mlsanity.com/paper/2409.16235/) (2024)
- [Under the Hood: Using Diagnostic Classifiers to Investigate and Improve how Language Models Track Agreement Information](https://mlsanity.com/paper/1808.08079/) (2018)
- [Testing the Predictions of Surprisal Theory in 11 Languages](https://mlsanity.com/paper/2307.03667/) (2023)
- [MultiBLiMP 1.0: A Massively Multilingual Benchmark of Linguistic Minimal Pairs](https://mlsanity.com/paper/2504.02768/) (2025)
- [As Good as New. How to Successfully Recycle English GPT-2 to Make Models for Other Languages](https://mlsanity.com/paper/2012.05628/) (2020)
- [CausalGym: Benchmarking causal interpretability methods on linguistic tasks](https://mlsanity.com/paper/2402.12560/) (2024)
- [Quantifying Context Mixing in Transformers](https://mlsanity.com/paper/2301.12971/) (2023)
- [Large Language Models Are Human-Like Internally](https://mlsanity.com/paper/2502.01615/) (2025)
- [Accounting for Agreement Phenomena in Sentence Comprehension with Transformer Language Models: Effects of Similarity-based Interference on Surprisal and Attention](https://mlsanity.com/paper/2104.12874/) (2021)
- [Causal Interventions Reveal Shared Structure Across English Filler-Gap Constructions](https://mlsanity.com/paper/2505.16002/) (2025)
- [Do LLMs dream of elephants (when told not to)? Latent concept association and associative memory in transformers](https://mlsanity.com/paper/2406.18400/) (2024)
- [Entropy- and Distance-Based Predictors From GPT-2 Attention Patterns Predict Reading Times Over and Above GPT-2 Surprisal](https://mlsanity.com/paper/2212.11185/) (2022)
- [Language Resources for Dutch Large Language Modelling](https://mlsanity.com/paper/2312.12852/) (2023)
- [NRGPT: An Energy-based Alternative for GPT](https://mlsanity.com/paper/2512.16762/) (2025)
- [MeMo: Towards Language Models with Associative Memory Mechanisms](https://mlsanity.com/paper/2502.12851/) (2025)
- [GEITje 7B Ultra: A Conversational Model for Dutch](https://mlsanity.com/paper/2412.04092/) (2024)
- [Diverging Transformer Predictions for Human Sentence Processing: A Comprehensive Analysis of Agreement Attraction Effects](https://mlsanity.com/paper/2603.16574/) (2026)

## Abstract (arXiv)

Psycholinguistics studies show that human readers fall for coherence illusions: an incoherent discourse can seem coherent simply because a distractor matches what comes next. We investigate whether Dutch language models (6 monolingual and 4 multilingual) show the same behavior on texts that link back to earlier context with words such as 'again' and 'too'. First, we find that surprisal at the critical word tracks human acceptability judgments and eye-tracking data. Models are more surprised by incoherent continuations, but a matching distractor in the prior context reduces this surprisal. Second, attention entropy identifies heads that behave differently under coherence vs. incoherence. We find that ablating these heads shows transfer effects across experiments, suggesting a shared mechanism. Third, we introduce energy from the associative-memory literature as a metric to quantify discourse coherence. Taken together, our results show that coherence illusions arise in Dutch LLMs, with entropy and energy exposing mechanisms that operate across settings.