--- title: "Syntactic Priming in Few-Shot Learning — Yadla 2026" description: "Few-shot prompting research has mostly examined the meaning, labels and reasoning steps of demonstrations." authors: ["Prasanth Yadla"] year: 2026 venue: "Proceedings of the 15th Joint Conference on Lexical and Computational Semantics (*SEM 2026)" source: "https://aclanthology.org/2026.starsem-conference.2/" doi: "10.18653/v1/2026.starsem-conference.2" licence: "CC BY 4.0 (ACL Anthology)" licence_url: "https://creativecommons.org/licenses/by/4.0/" summary_licence: "Elamur Cosmic License" canonical: "https://research.elamur.ai/free/en/papers/yadla-2026-syntactic-priming-few-shot-learning/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/yadla-2026-syntactic-priming-few-shot-learning/) [home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/9/) / Syntactic Priming in Few-Shot Learning: How Demonstration Structure Shapes LLM Performance # Syntactic Priming in Few-Shot Learning: How Demonstration Structure Shapes LLM Performance **Authors:** Prasanth Yadla · **Year:** 2026 · **Venue:** Proceedings of the 15th Joint Conference on Lexical and Computational Semantics (\*SEM 2026) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/) **Links:** [Publisher's page](https://aclanthology.org/2026.starsem-conference.2/) · [PDF](https://aclanthology.org/2026.starsem-conference.2.pdf) · [DOI 10.18653/v1/2026.starsem-conference.2](https://doi.org/10.18653/v1/2026.starsem-conference.2) · [OpenAlex](https://openalex.org/W7166838701) **Citations:** 0 citations (openalex-snapshot · 2026-10-14) ## Summary Few-shot prompting research has mostly examined the meaning, labels and reasoning steps of demonstrations. This paper asks whether their grammatical form also matters. Psycholinguistics shows that people tend to reuse sentence structures they have just encountered, a phenomenon called structural priming. The author tests whether LLMs do the same during in-context learning and what that means for task outputs. Five models from four families (Llama, Mistral, Qwen, Gemma), ranging from 7B to 14B parameters, were run locally via Ollama. They were prompted with synthetic, controlled English minimal pairs covering four alternations: active/passive, canonical/it-cleft, double-object/prepositional dative and particle placement. The prompts spanned three generation tasks: sentence completion, paraphrase and story continuation. Models reliably produced more of whichever variant appeared in the demonstrations. This held even in paraphrase settings where a different form was called for, and the effect lasted across several intervening sentences. The practical lesson is that demonstration syntax is not neutral. It is a separate design factor for prompt engineering and a possible confound in benchmark comparisons. The stimuli are released as the SyntaxPrime-ICL benchmark. ## Key points - Priming strength, measured as the ratio of target-construction production after matching versus opposing primes, ranged from roughly 1.3× to 6.4× depending on the construction. - Larger models tended to prime more strongly (r = 0.85, p = 0.068). This is a positive trend across 7B–14B models, but it is not significant at the conventional 0.05 level. - The effects were construction-specific rather than general stylistic drift: - Cleft and particle constructions almost never appeared under other prime types (rates of 0.00–0.04). - Voice and dative variants appeared often regardless of prime type (0.44–0.80), reflecting their high baseline frequency. - The study is limited to English, four alternations and generation tasks only. The lexical-overlap manipulation varied mainly verb overlap. The stimuli are synthetic minimal pairs rather than naturalistic prompts, and no causal or mechanistic analysis was performed. - Released resource: the SyntaxPrime-ICL benchmark, together with code and the experimental framework. *Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.* Fields: [Linguistics & languages](https://md.research.elamur.ai/free/en/shelves/linguistics/) Topics: [Language models & pretraining](https://md.research.elamur.ai/free/en/topics/language-models/) ## Cite The paper itself, not this page. BibTeX from the library's record of the work. ```bibtex @inproceedings{yadla2026syntactic, title = {{Syntactic Priming in Few-Shot Learning: How Demonstration Structure Shapes LLM Performance}}, author = {Yadla, Prasanth}, booktitle = {Proceedings of the 15th Joint Conference on Lexical and Computational Semantics (*SEM 2026)}, year = {2026}, doi = {10.18653/v1/2026.starsem-conference.2}, url = {https://doi.org/10.18653/v1/2026.starsem-conference.2} } ``` - [← previous SWAN: Semantic Watermarking with Abstract Meaning Representation](https://md.research.elamur.ai/free/en/papers/ye-2026-swan-semantic-watermarking-abstract-meaning/) - [next → Syntax is the Key to Semantics: Combining Universal Dependencies and Abstract Meaning Representation](https://md.research.elamur.ai/free/en/papers/heinecke-2026-syntax-is-key-semantics-combining/) - [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/9/) ## References 1 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. J. Kathryn Bock. 1986. Syntactic persistence in language production. Cognitive Psychology, 18(3):355– 387. \#\# Ethics Statement [find](https://scholar.google.com/scholar?q=J%2E+Kathryn+Bock%2E+1986%2E+Syntactic+persistence+in+language+production%2E+Cognitive+Psychology%2C+18%283%29%3A355%E2%80%93+387%2E+%5C%23%5C%23+Ethics+Statement)