# Limits of n-gram Style Control for LLMs via Logit-Space Injection

arXiv: https://arxiv.org/abs/2601.16224 · HTML: https://mlsanity.com/paper/2601.16224/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 7 of its 8 references that are in the map.
- First posted on arXiv 12 Jan 2026.

## Builds on (most cited first, 7 of 7)

- [Plug and Play Language Models: A Simple Approach to Controlled Text Generation](https://mlsanity.com/paper/1912.02164/) (2019)
- [Infini-gram: Scaling Unbounded n-gram Language Models to a Trillion Tokens](https://mlsanity.com/paper/2401.17377/) (2024)
- [Decoding-time Realignment of Language Models](https://mlsanity.com/paper/2402.02992/) (2024)
- [PAD: Personalized Alignment of LLMs at Decoding-Time](https://mlsanity.com/paper/2410.04070/) (2024)
- [Critic-Guided Decoding for Controlled Text Generation](https://mlsanity.com/paper/2212.10938/) (2022)
- [N-gram Is Back: Residual Learning of Neural Text Generation with n-gram Language Model](https://mlsanity.com/paper/2210.14431/) (2022)
- [Transferring Extreme Subword Style Using Ngram Model-Based Logit Scaling](https://mlsanity.com/paper/2503.08550/) (2025)

## Built on by (most cited first, 1 of 1)

- [Automated IEP Generation from Traditional Chinese Parent-Teacher Interviews via Corpus-Grounded Feature Diffusion](https://mlsanity.com/paper/2606.09603/) (2026)

## Abstract (arXiv)

Large language models (LLMs) are typically personalized via prompt engineering or parameter-efficient fine-tuning such as LoRA. However, writing style can be difficult to distill into a single prompt, and LoRA fine-tuning requires computationally intensive training and infrastructure. We investigate a possible lightweight alternative: steering a frozen LLM with n-gram style priors injected in logit space at decoding time. We train an n-gram model on stylistically distinct corpora -- including Don Quixote, CNN/DailyMail news headlines, and arXiv abstracts -- constructing an interpolated 1-to-3-gram prior over next-token probabilities. During generation we modify the LLM's logits by adding a weighted sum of style log-probabilities from each n-gram order that matches the current context, scaled by a control parameter lambda in [0, 1]. We sweep lambda and style corpora and report style perplexity under the n-gram model, base-model perplexity as a proxy for fluency, Jensen-Shannon (JS) divergence between the original and steered token distributions, and token-overlap statistics. On TinyLlama-1.1B we identify a single narrow regime (for the Don Quixote corpus at lambda=0.1) where style perplexity improves by 24.7% and base-model perplexity improves by 51.4% relative to the frozen model. Outside this regime, and for multi-author corpora such as CNN/DailyMail and arXiv abstracts, even small nonzero lambda values generally result in worse style and fluency, and larger lambda values lead to collapse with extreme perplexities and incoherent text. Logit-space injection of n-gram style priors provides lightweight, tunable style control, but it is fragile: it operates effectively only within a narrow range of low lambda values and is consistently outperformed by prompting and LoRA.