Language models retain ghost parses from garden-path sentences
Source published ·Modelwire updated
Original coverage: arXiv cs.CL ↗·How Modelwire adds context
The development
Researchers tested whether large language models retain abandoned syntactic analyses after encountering garden-path sentences, a classic psycholinguistic phenomenon. Across 24 models ranging from 124M to 70B parameters, they found persistent traces of incorrect parses that influence subsequent text generation, with residual effects reaching roughly 75% of the priming strength observed in humans. This work reveals how model internals preserve and compete between interpretations, offering insights into both model cognition and potential failure modes in complex linguistic reasoning.
Modelwire’s AI-generated summary of coverage from arXiv cs.CL.
Modelwire analysis
ExplainerOur AI-generated reading of the wider context and the next developments to watch.
The key finding isn't just that models retain wrong parses, but that these ghost interpretations actively compete with correct ones during generation. This reveals a mechanism, not just a failure: models don't erase bad hypotheses; they carry multiple interpretations forward and let them influence what comes next.
This connects directly to the October 22 work on state integrity collapse in long-horizon dialogues, which found that models retain constraints in memory yet stop enforcing them. Here we see a related phenomenon at the syntactic level: information persists internally even after being contradicted, shaping behavior downstream. Both papers expose a gap between what models store and what they apply. The garden-path work also echoes the perturbation study from October 23, which mapped how competing internal representations drive failures. Together these three papers suggest model fragility stems partly from unresolved competition between interpretations rather than simple forgetting.
If researchers can show that pruning or suppressing these residual parses during inference reduces downstream errors on complex sentences, that would confirm these traces are genuinely harmful rather than benign artifacts. Watch for follow-up work testing whether targeted intervention on abandoned analyses improves performance on tasks requiring robust syntactic reasoning.
This interpretation is generated from the summary above and the archive coverage cited below. Our methodology · Report an error
Coverage behind this analysis
These archive entries ground the connection in our analysis. They are ordered by source publication date, with links to our coverage and the original sources.
·arXiv cs.CL
LLMs retain rules but stop following them in long conversations
Researchers have identified a critical failure mode in long-context LLM conversations where models retain stated constraints in memory yet stop enforcing them over multiple turns. The new StateTrack-Bench evaluation reveals that compliance retention ranges from 62% to 100% across model snapshots, with failures concentrated in symbolic constraints rather than genuine forgetting. This behavioral drift, affecting…
MentionsarXiv
How this coverage is produced
Modelwire uses AI to generate summaries and context from source headlines, snippets, and selected archive coverage. Automated checks do not verify every claim, and items are not routinely reviewed by a person before publication. Zacaria Solis operates the site. Read the linked source for the full evidence and report errors through our corrections process.
Modelwire summarizes, we don’t republish. arXiv cs.CL originally reported this story as “Language Models Keep Abandoned Analyses: Residue and Adaptation after Garden-Path Sentences”. The full content lives on arxiv.org. If you’re a publisher and want a different summarization policy for your work, see our takedown page.