Priors Persist Through Suppression: A Stroop Paradigm for Lexical Override

• URL: http://arxiv.org/abs/2606.07555v1
• Date: Mon, 25 May 2026 05:53:31 GMT
• Title: Priors Persist Through Suppression: A Stroop Paradigm for Lexical Override
• Abstract summary: Glossaries, technical specifications, and system prompts routinely ask language models to use familiar words in unfamiliar ways.<n>When this works, the lexical prior persists through override rather than being replaced.<n>We test this with a Stroop-style paradigm: a remapping rule pitted against the query word's lexical-prioror ("hospital"), with matched neutral controls.
• Score: 2.892857112058156
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Glossaries, technical specifications, and system prompts routinely ask language models to use familiar words in unfamiliar ways. When this works, the lexical prior persists through override rather than being replaced: it continues to operate after the local rule applies, with the rule lowering its logit rather than installing the new meaning on top. We test this with a Stroop-style paradigm: a remapping rule ("doctor" means "forest") pitted against the query word's lexical-prior distractor ("hospital"), with matched neutral controls. Across 11 open-weight models spanning four families and 1B--9B parameters, lexical-prior strength predicts interference even after item-level controls for answer prior, frequency, tokenization, and prompt wording. Activation patching on five aligned models locates a source-position triplet (definition subject, definition target, query word) that nearly fully recovers the conflict effect (aggregate $R \in [0.92, 1.06]$). A definition-target swap shows the triplet performs binding rather than identity matching. Dissociation experiments isolate target preservation as the binding-specific signature: distractor suppression occurs under matched, swap, and item-mismatched conditions alike, whereas target logit collapse occurs only when the definition-target position is corrupted. Behavior and mechanism converge on the same channel: the lexical prior is where both interference originates and where override leaves its mark.

Related papers

• A Dynamic-Semantics Framework for Grounding Human Referring Expressions in Visual Perceptual Data [0.0]
  We present a framework that externalizes pact state into three explicit, inspectable sets of referent-object bindings.<n>We evaluate the framework on the Stanford Repeated Reference Game corpus.
  arXiv  Detail & Related papers  (2026-08-09T12:14:10Z)
• Trajectories That Segment Themselves: Agent-Declared Boundaries as a Training Unit [0.0]
  We introduce collection-time semantic self-segmentation, in which a declarative contract has the acting agent expose its own boundaries while the trajectory is generated.<n>Because the agent names its conjecture, a reviewer can negate it by name, which lets our protocol manufacture wrong-cause-then-correction transitions that recorded work rarely contains.<n>We then ask what survives deleting the declaration. Given the cut points but not the hypothesis, a model attributes action blocks to their governing hypothesis at over twice chance.<n>Asked instead to place boundaries, a code-blind annotator matches 24 of 40 where random placement matches 11.5, while a mechanical test
  arXiv  Detail & Related papers  (2026-08-03T14:27:58Z)
• Grounding Spatial Relations in a Compact World Model: Instruction Leakage and a Goal-Free Dynamics Fix [46.122203287541005]
  We show that a goal-conditioned predictor reaches a striking $0.90$ relation-readout accuracy, yet this is emphinstruction transcription, not perception.<n>Withholding the goal collapses it to chance ($0.90!to!0.27$, three seeds) and a counterfactual instruction makes the predicted anchors follow the emphfalse instruction $94.5%$ of the time.<n>The diagnosis prescribes the fix: keep the goal out of the dynamics (it belongs to the planner's cost) and supervise the emphread path
  arXiv  Detail & Related papers  (2026-07-08T02:38:43Z)
• Weave of Formal Thought [51.56484100374058]
  We introduce Weave of Formal Thought (WoFT), a paradigm uniting rigorous syntactic validation with learned structural representations.<n>We present a latent-variable fine-tuning method training the language model to interleave non-terminal grammar symbols directly into generation.<n>For Python, fine-tuning StarCoder2-3B with our RWS objective reduces per-token cross-entropy by 14.3% relative to a text-only SFT baseline.
  arXiv  Detail & Related papers  (2026-06-24T15:58:11Z)
• When to Think, When to Speak: Learning Disclosure Policies for LLM Reasoning [60.714416943748866]
  In single-stream autoregressive interfaces, the same tokens both update the model state and constitute an irreversible public commitment.<n>We introduce Side-by-Side (SxS) Interleaved Reasoning, which makes disclosure timing a controllable decision within standard autoregressive generation.
  arXiv  Detail & Related papers  (2026-05-05T02:59:58Z)
• Beyond Perception Errors: Semantic Fixation in Large Vision-Language Models [1.7767466724342065]
  Large vision-language models (VLMs) often rely on familiar semantic priors, but existing evaluations do not cleanly separate perception failures from rule-mapping failures.<n>We study this behavior as semantic fixation: preserving a default interpretation even when the prompt specifies an alternative, equally valid mapping.<n>Across 14 open and closed VLMs, accuracy consistently favors standard rules, revealing a robust semantic-fixation gap.
  arXiv  Detail & Related papers  (2026-04-13T22:53:39Z)
• Triangulation as an Acceptance Rule for Multilingual Mechanistic Interpretability [0.0]
  We argue that mechanistic explanations for such models should satisfy a emphcausal standard.<n>Claims must survive causal interventions and must emphcross-reference across environments that perturb surface form while preserving meaning.<n>We ground triangulation in causal abstraction by casting it as an approximate transformation score over a distribution of interchange interventions, connect it to the pragmatic interpretability agenda, and present a comparative experimental protocol across model families, language pairs, and tasks.
  arXiv  Detail & Related papers  (2025-12-31T13:03:34Z)
• Universal Adversarial Suffixes Using Calibrated Gumbel-Softmax Relaxation [9.099589602551573]
  We study universal adversarial suffixes that, when appended to any input, broadly reduce accuracy across tasks and models.<n>Our approach learns the suffix in a differentiable "soft" form using Gumbel-Softmax relaxation and then discretizes it for inference.<n>A single suffix trained on one model transfers effectively to others, consistently lowering both accuracy and calibrated confidence.
  arXiv  Detail & Related papers  (2025-12-09T00:03:39Z)
• MaxSup: Overcoming Representation Collapse in Label Smoothing [52.66247931969715]
  Label Smoothing (LS) is widely adopted to reduce overconfidence in neural network predictions.<n>LS compacts feature representations into overly tight clusters, diluting intra-class diversity.<n>We propose Max Suppression (MaxSup), which applies uniform regularization to both correct and incorrect predictions.
  arXiv  Detail & Related papers  (2025-02-18T20:10:34Z)
• Definition generation for lexical semantic change detection [3.7297237438000788]
  We use contextualized word definitions generated by large language models as semantic representations in the task of diachronic lexical semantic change detection (LSCD) In short, generated definitions are used as senses', and the change score of a target word is retrieved by comparing their distributions in two time periods under comparison. Our approach is on par with or outperforms prior non-supervised LSCD methods.
  arXiv  Detail & Related papers  (2024-06-20T10:13:08Z)
• Biomedical Named Entity Recognition via Dictionary-based Synonym Generalization [51.89486520806639]
  We propose a novel Synonym Generalization (SynGen) framework that recognizes the biomedical concepts contained in the input text using span-based predictions. We extensively evaluate our approach on a wide range of benchmarks and the results verify that SynGen outperforms previous dictionary-based models by notable margins.
  arXiv  Detail & Related papers  (2023-05-22T14:36:32Z)
• Efficient Zero-shot Event Extraction with Context-Definition Alignment [50.15061819297237]
  Event extraction (EE) is the task of identifying interested event mentions from text. We argue that using the static embedding of the event type name might not be enough because a single word could be ambiguous. We name our approach Zero-shot Event extraction with Definition (ZED)
  arXiv  Detail & Related papers  (2022-11-09T19:06:22Z)
• Connect-the-Dots: Bridging Semantics between Words and Definitions via Aligning Word Sense Inventories [47.03271152494389]
  Word Sense Disambiguation aims to automatically identify the exact meaning of one word according to its context. Existing supervised models struggle to make correct predictions on rare word senses due to limited training data. We propose a gloss alignment algorithm that can align definition sentences with the same meaning from different sense inventories to collect rich lexical knowledge.
  arXiv  Detail & Related papers  (2021-10-27T00:04:33Z)
• Semantic-Preserving Adversarial Text Attacks [85.32186121859321]
  We propose a Bigram and Unigram based adaptive Semantic Preservation Optimization (BU-SPO) method to examine the vulnerability of deep models. Our method achieves the highest attack success rates and semantics rates by changing the smallest number of words compared with existing methods.
  arXiv  Detail & Related papers  (2021-08-23T09:05:18Z)
• Adversarial Semantic Collisions [129.55896108684433]
  We study semantic collisions: texts that are semantically unrelated but judged as similar by NLP models. We develop gradient-based approaches for generating semantic collisions. We show how to generate semantic collisions that evade perplexity-based filtering.
  arXiv  Detail & Related papers  (2020-11-09T20:42:01Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.