A measurement instrument · EMNLP / ACL / arXiv
Measuring human rhythm in language generation.
Burstiness is the variation in sentence length and complexity that makes writing feel human. Its working definition today is an unreviewed detector heuristic. We replace it with a reproducible measurement instrument, and benchmark how far prompting moves it across nine model families.
METAVENTIONS AI · Research preprint in progress
64 papers
4 research passes
214 tests passing
Boundary dial: monotone ρ = 1.0, robustly directional p < 10-5 · coherence held
Human rhythmhigh var(L)
Machine textflat
Same content, two rhythms. We give the difference a definition that is reproducible, scale-free, and model-free.
Abstract
A property to be specified and steered, not a heuristic to be measured.
Burstiness is the most reliable signal that text was produced by a language model, yet it is defined only by a detector heuristic and controlled only by post-hoc rewriting. We treat it instead as a property that can be specified as a distributional target and steered during generation.
We frame the work as modeling human rhythmic patterns, not as evading detection, and we report detector behavior across the full control range so the contribution serves measurement and model analysis as well as generation. The novelty is the composition: applying established steering machinery to a rhythm-specific target, with a reusable definition and perceptual validation.
The wedge
The bridge between two camps is empty.
Demand for human rhythm in AI text is everywhere: detectors, humanizer tools, prompt tricks. The literature splits cleanly, and nobody has crossed the gap.
A
Measurement and detection
A strong, recent line of work quantifies burstiness after the fact to catch machine text. None of it controls the property.
B
Steering machinery
Activation and LoRA methods steer frozen models toward style targets: sentiment, persona, authorship. None of them target rhythm.
→
Our contribution
A rhythm-specific distributional target, steered in-generation, validated by human perception. The composite is the novelty, not any single mechanism.
A reproducible definition
Burstiness as a vector, not a single scalar.
The canonical operational definition of burstiness in detection practice is a blog post. The field runs on an un-formalized metric. We replace it with a decomposable, reproducible target.
The burstiness vector
B = [ var(L), kurt(L), mean_surprisal, fluc_dabs(S), punct_entropy ]
Variance and kurtosis of sentence length, surprisal fluctuation under a fixed reference model, and punctuation entropy. Our ablation shows the field's implicit metric, standard deviation of surprisal, is the weakest discriminator. Local jumpiness and mean surprisal separate human from machine far better.
Four contributions
From a heuristic to a controller.
01
A reproducible definition
Burstiness as a decomposable distributional target, replacing the single-scalar detector heuristic.
02
The prompt-control ceiling
A clean measurement of how prompt-only control degrades against the target over output length. A standalone negative result.
03
An in-generation controller
Biasing only the sentence-boundary decision yields a monotonic, coherence-preserving length dial. On gpt2-medium it is monotone (Spearman ρ 1.0), coherence-preserving, and robustly controllable in direction: 40 of 48 matched pairs respond as commanded (sign and Wilcoxon p < 10-5). The residual is magnitude, not existence, since var(L) is heavy-tailed and mean-based effect sizes understate it.
04
A perception protocol
A pre-registered human study, adapted from a method validated in speech synthesis, linking controlled rhythm to perceived humanness.
The lab
Where the systems converge.
The Burstiness Engine is built inside METAVENTIONS AI as a self-rebuilding research system: a curated literature corpus, an experiment harness, and a paper that regenerates from a single source of truth. Two people, one rhythm.
VL
Vittoria Lanzo
Co-author
Owns the research spine and the perception protocol: prompt-level control is the baseline, model-level steering is the answer.
Burstiness Engine
The research questions
Seven questions, mapped to the literature.
| # | Question | Status |
|---|---|---|
| Q1 | The prompt-engineering ceiling for rhythmic control | Gap, publishable |
| Q2 | Burstiness as a learnable rhythm embedding, no base retrain | Mechanism exists |
| Q3 | Is AI text getting more bursty across generations? | Empirically yes |
| Q4 | Small auxiliary rhythm model, a burstiness LoRA | Controller in progress |
| Q5 | Sentence variance vs inter-token timing | Variance formalized |
| Q6 | Burstiness fingerprint vs idiosyncratic style | Stylometry strong |
| Q7 | Punctuation and paragraph structure as proxies | Genuine gap |
A deeper walk through the four research passes and the methodology lives on the research page.
Current status
The framing, definition, ablation, negative result, and a robustly directional controller are done. The open frontier is a clean magnitude headline at a larger base. Explore the live corpus dashboard or the results and figures.