Do Language Models Know What Not to Say? Causal Evidence for Statistical Preemption in LLMs
A recent study explores how language models learn what not to say through statistical preemption. The research demonstrates that these models can acquire negative linguistic knowledge by competing with alternative forms. Findings indicate that model size influences preemption sensitivity and that manipulating competing-form frequencies can alter preemption behavior.
- ▪The study investigates the concept of statistical preemption in large language models.
- ▪Results show a strong correlation between LLM surprisal patterns and human acceptability judgments.
- ▪The research confirms that competing-form frequency drives preemption sensitivity rather than overall verb frequency.
Opening excerpt (first ~120 words) tap to expand
Computer Science > Computation and Language arXiv:2605.23039 (cs) [Submitted on 21 May 2026] Title:Do Language Models Know What Not to Say? Causal Evidence for Statistical Preemption in LLMs Authors:Dongxin Guo, Jikun Wu, Siu Ming Yiu View a PDF of the paper titled Do Language Models Know What Not to Say? Causal Evidence for Statistical Preemption in LLMs, by Dongxin Guo and 2 other authors View PDF HTML (experimental) Abstract:How do learners acquire knowledge of what is unacceptable without negative evidence? Construction Grammar proposes statistical preemption: exposure to a conventional form (e.g., "donated the books to the library") preempts structurally possible but unattested alternatives ("*donated the library the books").
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at arXiv cs.AI.
Discussion
0 commentsMore from arXiv cs.AI
