Priors Persist Through Suppression: A Stroop Paradigm for Lexical Override

arXiv:2606.07555v2 Announce Type: replace-cross Abstract: Glossaries, technical specifications, and system prompts routinely ask language models to use familiar words in unfamiliar ways. When this works, the local rule does not install the new meaning on top of the old one; the pretrained prior keeps operating underneath, and its strength still shows through. 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 prior's strength both predicts which overrides fail and marks where the causal repair lands.

Abstract:Glossaries, technical specifications, and system prompts routinely ask language models to use familiar words in unfamiliar ways. When this works, the local rule does not install the new meaning on top of the old one; the pretrained prior keeps operating underneath, and its strength still shows through. 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 prior's strength both predicts which overrides fail and marks where the causal repair lands.

Computation and Language (cs.CL); Machine Learning (cs.LG)

arXiv:2606.07555 [cs.CL]

(or arXiv:2606.07555v2 [cs.CL] for this version)

https://doi.org/10.48550/arXiv.2606.07555 Focus to learn more arXiv-issued DOI via DataCite

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.

aiqbench analysis

Priors Persist Through Suppression: A Stroop Paradigm for Lexical Override

Related coverage