# The Dog the Cat Chased Stumped the Model: Measuring When Language Models Abandon Structure for Shortcuts

arXiv: https://arxiv.org/abs/2510.20543 · HTML: https://mlsanity.com/paper/2510.20543/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 18 of its 44 references that are in the map.
- First posted on arXiv 23 Oct 2025.

## Builds on (most cited first, 18 of 18)

- [Measuring Massive Multitask Language Understanding](https://mlsanity.com/paper/2009.03300/) (2020)
- [MiniLM: Deep Self-Attention Distillation for Task-Agnostic Compression of Pre-Trained Transformers](https://mlsanity.com/paper/2002.10957/) (2020)
- [Humanity's Last Exam](https://mlsanity.com/paper/2501.14249/) (2025)
- [Entity-Based Knowledge Conflicts in Question Answering](https://mlsanity.com/paper/2109.05052/) (2021)
- [Reasoning or Reciting? Exploring the Capabilities and Limitations of Language Models Through Counterfactual Tasks](https://mlsanity.com/paper/2307.02477/) (2023)
- [A Systematic Assessment of Syntactic Generalization in Neural Language Models](https://mlsanity.com/paper/2005.03692/) (2020)
- [DisentQA: Disentangling Parametric and Contextual Knowledge with Counterfactual Question Answering](https://mlsanity.com/paper/2211.05655/) (2022)
- [Reevaluating Adversarial Examples in Natural Language](https://mlsanity.com/paper/2004.14174/) (2020)
- [Which of These Best Describes Multiple Choice Evaluation with LLMs? A) Forced B) Flawed C) Fixable D) All of the Above](https://mlsanity.com/paper/2502.14127/) (2025)
- [Mechanisms for Handling Nested Dependencies in Neural-Network Language Models and Humans](https://mlsanity.com/paper/2006.11098/) (2020)
- [A Targeted Assessment of Incremental Processing in Neural LanguageModels and Humans](https://mlsanity.com/paper/2106.03232/) (2021)
- [A Glitch in the Matrix? Locating and Detecting Language Model Grounding with Fakepedia](https://mlsanity.com/paper/2312.02073/) (2023)
- [In-context Learning Generalizes, But Not Always Robustly: The Case of Syntax](https://mlsanity.com/paper/2311.07811/) (2023)
- [Plausibly Problematic Questions in Multiple-Choice Benchmarks for Commonsense Reasoning](https://mlsanity.com/paper/2410.10854/) (2024)
- [Hierarchical Representation in Neural Language Models: Suppression and Recovery of Expectations](https://mlsanity.com/paper/1906.04068/) (2019)
- [On the Empirical Complexity of Reasoning and Planning in LLMs](https://mlsanity.com/paper/2404.11041/) (2024)
- [Don't Search for a Search Method -- Simple Heuristics Suffice for Adversarial Text Attacks](https://mlsanity.com/paper/2109.07926/) (2021)
- [Fine-Tuned LLMs are "Time Capsules" for Tracking Societal Bias Through Books](https://mlsanity.com/paper/2502.05331/) (2025)

## Abstract (arXiv)

When language models correctly parse "The cat that the dog chased meowed," are they analyzing syntax or simply familiar with dogs chasing cats? Despite extensive benchmarking, we lack methods to distinguish structural understanding from semantic pattern matching. We introduce CenterBench, a dataset of 9,720 comprehension questions on center-embedded sentences (like "The cat [that the dog chased] meowed") where relative clauses nest recursively, creating processing demands from simple to deeply nested structures. Each sentence has a syntactically identical but semantically implausible counterpart (e.g., mailmen prescribe medicine, doctors deliver mail) and six comprehension questions testing surface understanding, syntactic dependencies, and causal reasoning. Testing six models reveals that performance gaps between plausible and implausible sentences widen systematically with complexity, with models showing median gaps up to 26.8 percentage points, quantifying when they abandon structural analysis for semantic associations. Notably, semantic plausibility harms performance on questions about resulting actions, where following causal relationships matters more than semantic coherence. Reasoning models improve accuracy but their traces show semantic shortcuts, overthinking, and answer refusal. Unlike models whose plausibility advantage systematically widens with complexity, humans shows variable semantic effects. CenterBench provides the first framework to identify when models shift from structural analysis to pattern matching.