Elevated detector scores track long-token and Academic Word List density, with numeric and non-alphabetic token ratios correlating negatively¶
Claim · All claims
Evidence · 1 study · causalr2·q2·i2medium
Currency · newest study 2026 · not yet checked against its sources · not reviewed by a person · recheck by 2029: tool and model accuracy changes quickly
Subclaims¶
Reading the evidence codes
q is the design tier, i the impact, n the sample, and the kind with its r
the rigour judged within that kind: what each code means.
This claim: no standardized effect is recorded in observations/ for it yet, so no benchmark reading is possible.
q2 i2 Long-token and AWL ratios correlate positively with 2023–2025 detector scores (ρ ≈ 0.30 to 0.35, p < 0.001); numeric and non-alphabetic ratios correlate negatively. → Jonathan A. Karr Jr. 2026
q2 i2 After domain centering, long-token and AWL associations strengthen for Pangram (domain-adjusted r > 0.41), indicating response to within-domain stylistic cues. → Jonathan A. Karr Jr. 2026
Evidence¶
Jonathan A. Karr Jr. 2026¶
Jonathan A. Karr Jr., Grigorii Khvatskii, Ting Hua, and Nitesh V. Chawla. (2026). Why AI Detection Fails for Academic Integrity. Proceedings of the ACM AI Leadership Summit (AILS '26). https://arxiv.org/abs/2608.11256 doi:10.48550/arxiv.2608.11256
q2 · i2 · causal · r2
Correlational analysis (Spearman ρ with Benjamini-Hochberg FDR) of token-level features against Pangram and GPTZero scores on 2023–2025 texts. The article reports "𝜌≈ 0.30to0 .35;𝑝< 0.001" for long-token and AWL ratios, with Pangram AWL ρ = 0.346.
"Long-token and AWL ratios correlate positively with 2023 to 2025 scores (𝜌≈ 0.30to0 .35;𝑝< 0.001); numeric and non-alphabetic ratios correlate negatively (Table 6; full tests in Appendix H)."
Discussion¶
Related Claims¶
- Humanization works by systematically suppressing detector-salient surface cues — long-token ratio, AWL density, and lexical diversity — rather than fragmenting syntax — related
- Detector flag rates on 2023–2025 original abstracts vary sharply by domain, with non-STEM rates far above STEM — related
- GPT-assisted writing shows higher lexical diversity, richness, and noun/non-stopword ratios, while independently authored writing shows higher adverb ratio and sentence-length variability — related
Page metadata
| Field | Value |
|---|---|
type |
claim |
id |
detector-scores-track-long-token-and-awl-density |
title |
Elevated detector scores track long-token and Academic Word List density, with numeric and non-alphabetic token ratios correlating negatively |
description |
Elevated detector scores track long-token and Academic Word List density, with numeric and non-alphabetic token ratios correlating negatively |
status |
review |
evidence_strength |
moderate |
newest_evidence |
2026 |
recheck_by |
2029 |
generated |
by process:wiki-ingest on 2026-10-11 |
trust |
unverified — no one has confirmed this page's content |
sources |
1 — jonathan-a-karr-jr-2026 |
Frontmatter is the page's machine-readable record — see the Schema & Guide.