aiqbench analysis
This research demonstrates a fundamental limitation in current language models where pretrained semantic priors persist even after successful instruction-based remapping. By applying a Stroop-style paradigm across multiple open-weight architectures, the study reveals that overriding word meanings does not suppress old associations but rather creates an asymmetric state protected by specific prompt positions. These findings are critical for understanding how to effectively manage lexical interference in future AI systems and refine alignment techniques beyond simple suppression strategies.
Priors Persist Through Suppression: A Stroop Paradigm for Lexical Override
arXiv:2606.07555v4 Announce Type: replace Abstract: Glossaries, technical specifications, and system prompts routinely ask language models to use familiar words in unfamiliar ways. The instruction competes with what the word already means, and even when it wins, the pretrained prior keeps operating underneath. We test this with a Stroop-style paradigm: a prompt redefines a word (doctor now means forest), asks for a related word, and we score the new meaning against the word's pretrained associate (hospital) under matched neutral controls. Across 11 open-weight models from 1B to 9B parameters, the old meaning interferes in every model, remapping type, and prompt framing we test. After item-level controls, a model's ordinary preference for the old associate predicts the size of the interference, in the three document-relevant remapping types though not in antonyms. On antonym remapping, activation patching in five models locates the repair: restoring three prompt positions (where the word is redefined, where its new meaning appears, and where the question repeats it) recovers almost all of the effect (normalized recovery R in [0.92, 1.06]). The repair is asymmetric. The old meaning's logit falls under any perturbation of those positions, so pushing it down is not what separates a working override from a failing one; the new meaning survives only while the position carrying it is intact. What a local definition achieves is a protected new meaning, not a suppressed old one.
Abstract:Glossaries, technical specifications, and system prompts routinely ask language models to use familiar words in unfamiliar ways. The instruction competes with what the word already means, and even when it wins, the pretrained prior keeps operating underneath. We test this with a Stroop-style paradigm: a prompt redefines a word (doctor now means forest), asks for a related word, and we score the new meaning against the word's pretrained associate (hospital) under matched neutral controls. Across 11 open-weight models from 1B to 9B parameters, the old meaning interferes in every model, remapping type, and prompt framing we test. After item-level controls, a model's ordinary preference for the old associate predicts the size of the interference, in the three document-relevant remapping types though not in antonyms. On antonym remapping, activation patching in five models locates the repair: restoring three prompt positions (where the word is redefined, where its new meaning appears, and where the question repeats it) recovers almost all of the effect (normalized recovery R in [0.92, 1.06]). The repair is asymmetric. The old meaning's logit falls under any perturbation of those positions, so pushing it down is not what separates a working override from a failing one; the new meaning survives only while the position carrying it is intact. What a local definition achieves is a protected new meaning, not a suppressed old one.
Computation and Language (cs.CL); Machine Learning (cs.LG)
arXiv:2606.07555 [cs.CL]
(or arXiv:2606.07555v4 [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.
Full text under the publisher's open licence · 320 words
Related coverage
Priors Persist Through Suppression: A Stroop Paradigm for Lexical Override
arXiv · Jun 9, 2026
Priors Persist Through Suppression: A Stroop Paradigm for Lexical Override
arXiv · Jun 17, 2026
Reflections on unlearning and inoculation
LessWrong · Sep 20, 2026
MemVenom: Triggered Poisoning of Multimodal Memories in Web Agents
arXiv · Jun 10, 2026