Rhythm Variation
Analyzes sentence length patterns to detect unnaturally smooth rhythm. Human writing has variable sentence lengths that create a natural cadence, while AI tends to produce more uniform sentence lengths.
Technical description
Extracts sentence lengths (in words) and computes: coefficient of variation (CV) of sentence lengths, autocorrelation at lag 1-3 (do similar lengths cluster?), longest monotonic runs, and burstiness (ratio of variance to mean). Uses NLP sentence boundary detection for accurate segmentation. Low CV and high autocorrelation indicate the unnaturally smooth rhythm typical of AI text.
How it works
Layer 2 (NLP): Uses spaCy for accurate sentence boundary detection. Computes sentence length sequences. Calculates coefficient of variation and lag-based correlations. Detects monotonic runs (sequences of increasing or decreasing length). Measures burstiness of the length distribution. Low variation and high smoothness flag AI generation.
Why this matters
Human writers naturally vary sentence length — short sentences for emphasis, longer ones for complex ideas. This creates a distinctive rhythm. AI models tend to generate sentences of similar length with gradual variation, producing a smooth but unnatural cadence. The statistical properties of sentence length sequences can distinguish human from machine writing.
Score thresholds
- 0-1
- Natural rhythm with varied sentence lengths
- 2-3
- Somewhat smooth rhythm with reduced variation
- 4-5
- Unnaturally uniform sentence lengths throughout
Limitations
Technical writing naturally has more uniform sentence lengths. Very short texts provide insufficient data for rhythm analysis. Some literary styles deliberately use uniform sentence lengths for effect.