Tortured phrases: A dubious writing style emerging in science. Evidence of critical issues affecting established journals
A journal fast-tracked hundreds of papers with garbled tortured phrases, reused images, and abstracts a GPT detector flagged as synthetic.
2021/07/12 by Guillaume Cabanac, Cyril Labbé, Cabanac, Guillaume +3 · 15 voices · 12 citations
#cs.DL #cs.CL #cs.CY #cs.IR
paper · pdf · doi:10.48550/arxiv.2107.06751
Abstract
Probabilistic text generators have been used to produce fake scientific papers for more than a decade. Such nonsensical papers are easily detected by both human and machine. Now more complex AI-powered generation techniques produce texts indistinguishable from that of humans and the generation of scientific texts from a few keywords has been documented. Our study introduces the concept of tortured phrases: unexpected weird phrases in lieu of established ones, such as 'counterfeit consciousness' instead of 'artificial intelligence.' We combed the literature for tortured phrases and study one reputable journal where these concentrated en masse. Hypothesising the use of advanced language models we ran a detector on the abstracts of recent articles of this journal and on several control sets. The pairwise comparisons reveal a concentration of abstracts flagged as 'synthetic' in the journal. We also highlight irregularities in its operation, such as abrupt changes in editorial timelines. We substantiate our call for investigation by analysing several individual dubious articles, stressing questionable features: tortured writing style, citation of non-existent literature, and unacknowledged image reuse. Surprisingly, some websites offer to rewrite texts for free, generating gobbledegook full of tortured phrases. We believe some authors used rewritten texts to pad their manuscripts. We wish to raise the awareness on publications containing such questionable AI-generated or rewritten texts that passed (poor) peer review. Deception with synthetic texts threatens the integrity of the scientific literature.
Summary
The authors coin the term 'tortured phrases' for scientific terms garbled by word-for-word synonym substitution (e.g. 'counterfeit consciousness' for 'artificial intelligence') and trace them to one Elsevier journal, Microprocessors and Microsystems, where they concentrate. They show that journal's 2021 papers were peer-reviewed far faster than before, disproportionately came from mainland China, and had abstracts flagged as machine-generated by a GPT-2 detector at far higher rates than five comparison sets, and they document individual articles with reused images, fabricated citations, and unintelligible passages. They stop short of proving deliberate fraud and instead call on editors and publishers to investigate.
machine-generated · claude-sonnet-5
Outline
- Introduction — Frames tortured phrases within the history of publishing stings and paper mills, and previews the paper's investigative chain of evidence.
- Tortured phrases found in published academic articles — Defines the concept and lists ~25 examples found by manual search and snowballing, e.g. 'random forest' rendered as 'irregular timberland'.
- The Microprocessors and Microsystems journal — Profiles the journal (Elsevier, Q3 in Scimago) and describes the 1,078-article dataset built from Crossref DOIs and Elsevier's full-text API.
- Irregularities of the editorial assessment — Shows a 5-6x drop in submission-to-acceptance time starting in early 2021, a strong skew toward Chinese affiliations among fast-tracked papers, and blocks of papers sharing near-identical submission/revision/acceptance dates.
- Abstracts with high GPT detector score — Runs a RoBERTa-based GPT-2 output detector on 389 fast-tracked abstracts plus five control sets, using Dvoretzky-Kiefer-Wolfowitz confidence bands to show the journal's scores are significantly higher.
- Critical flaws found in questionable and problematic publications — Walks through six individual articles with unattributed image/text reuse, fabricated references, and passages the authors could not parse for meaning.
- Discussion and call for investigation — Argues these signals jointly point to misuse of automatic translation, rewriting, or generation tools and asks editors, publishers, and other researchers to investigate further.
machine-generated · claude-sonnet-5
Claims
- Scientific terms are being replaced in published papers by garbled 'tortured phrase' synonym substitutions (e.g. 'flag to clamor' for 'signal to noise'), found via literature search and full-text indexing. [experiment]
- In Microprocessors and Microsystems, the median time from submission to acceptance fell from 161 days (volumes 56-79, 2018-2020) to 25 days (volumes 80-83, early 2021), a roughly 6-fold decrease. [experiment]
- Papers accepted in under 30 days at this journal were overwhelmingly from mainland China (394/404, 97.5%), versus a small minority (58/615, 9.5%) among papers taking over 40 days. [experiment]
- Abstracts from the journal's fast-tracked 2021 articles score significantly higher on a GPT-2 synthetic-text detector than abstracts from the same journal's earlier volumes and four other control sets, with the difference confirmed via Dvoretzky-Kiefer-Wolfowitz confidence bands at 95% confidence. [experiment]
- 72.1% of the journal's 2021 articles had abstracts scoring 70% or higher on the GPT detector, far above the next-highest Elsevier journal in the comparison (13.6%). [experiment]
- Several individual articles in the journal contain images and passages copied from unrelated sources without attribution, citations to non-existent literature, and internal references (e.g. to a 'Theorem 1.2') that do not exist in the paper. [experiment]
machine-generated · claude-sonnet-5
Key figure
Figure 5 — Side-by-side comparison showing that a suspect article's figure and its surrounding paragraph were copied, with the wording heavily and confusingly altered, from an unrelated water-leak-detector circuit tutorial on a hobbyist electronics website.
machine-generated · claude-sonnet-5
Glossary
- Tortured phrase
- A standard technical term mangled by substituting each word with a synonym, producing an odd phrase in its place, e.g. 'counterfeit consciousness' instead of 'artificial intelligence.'
- Paper mill
- An organization that manufactures and sells fake or low-quality manuscripts, authorships, or citations to researchers under pressure to publish.
- GPT-2 Output Detector
- A RoBERTa-based classifier, released alongside GPT-2, that estimates the probability a given piece of text was machine-generated.
- Editorial assessment duration
- The time elapsed between a manuscript's submission and its acceptance, covering screening, review rounds, and revisions.
- Dvoretzky-Kiefer-Wolfowitz (DKW) inequality
- A statistical bound used to build a confidence band around a distribution estimated from sample data, letting the authors claim two groups' score distributions differ with stated confidence.
- Control set
- A comparison group of texts, unrelated to the journal under suspicion, used as a baseline to judge whether the suspicious group's scores are unusually high.
- PubPeer
- A website where readers publicly comment on published papers to flag errors, misconduct, or other concerns after publication.
- Retraction
- A publisher's formal withdrawal of a paper's validity due to serious errors or misconduct, while the paper typically remains online marked as retracted.
machine-generated · claude-sonnet-5
Audience
Journal editors, publishers, and peer reviewers responsible for catching fraudulent or synthetic submissions; researchers in research integrity, scientometrics, and misconduct detection; science journalists covering publishing scandals.
prerequisites: Basic familiarity with how peer review and journal publishing work, Some awareness of language models like GPT-2/GPT-3 and what 'synthetic text detection' means, Comfort reading simple summary statistics and confidence intervals, though not required to follow the main argument
machine-generated · claude-sonnet-5
Open questions
- What editorial process, if any, could produce dozens of papers sharing near-identical submission, revision, and acceptance dates?
The authors explicitly state they 'were not able to propose a satisfactory explanation ... within the bounds of a normal editorial process,' leaving open whether this reflects manipulated records, editorial overload, or something else. - Is the small sample size of control set D (50 Chinese-journal abstracts) responsible for its failure to separate cleanly from the experimental set at low GPT-score cutoffs, or does this reflect a genuine overlap between machine translation artifacts and the phenomenon under study?
The paper flags this as undecided and hypothesizes insufficient statistical power, which bears on whether translation alone (versus deliberate generation) can explain high detector scores. - What specific tool or process actually coins tortured phrases: faulty machine translation, an automatic paraphrasing/rewriting service, or full text generation?
The authors call the mechanism 'quite mysterious' and only hypothesize misuse of NLP tools, without pinning down which one or how it enters the manuscript pipeline. - Does the same pattern of fast-tracked review, high GPT-detector scores, and tortured phrases appear in other journals or publishers beyond the single Elsevier journal examined?
The paper is explicitly framed as 'an open call for investigation' inviting others to replicate the analysis elsewhere, rather than a claim about the scope of the problem.
machine-generated · claude-sonnet-5
Supplementary links
machine-generated · claude-sonnet-5
Cited by
Discussions
- A dubious writing style emerging in science [hn, 110 points, 40 comments]
- In 2021 Cabanac, Labbé and mathematician Alexander Magazinov had published "Tortured phrases: A dubious writing style emerging in science", introducing the world to thesaurus-hacked articles. It has [bsky, 39 points, 1 comments]
- Plenty of tortured phrases (arxiv.org/abs/2107.0...) in this #IEEE paper. ieeexplore.ieee.org/... My pubpeer report: pubpeer.com/publicat... [bsky, 26 points, 2 comments]
- CL: Later, 'tortured phrases' popped up, a way to avoid plagiarism detection. arxiv.org/abs/2107.06751 We also built a database of papers that contained such phrases and other fishy papers. www.irit.f [bsky, 10 points, 2 comments]
- Interesting study on tortured phrases as an indication of GPT/LLM usage in scientific journals. arxiv.org/abs/2107.06751 [bsky, 3 points, 2 comments]
- Background on tortured phrases like "counterfeit consciousness" in place of "artificial intelligence" or "bogus upsides" instead of "false positives": doi.org/10.48550/arX... [bsky, 3 points, 2 comments]
- #arXiv Tortured phrases: A dubious writing style emerging in science. Evidence of critical issues affecting established journals arxiv.org/abs/2107.06751 Detection of tortured phrases in scientific li [bsky, 3 points, 0 comments]
- Somewhat besides the point, but multiple plagiarists and generated papers have also gotten caught specifically *because* they tried to rephrase common terminology: arxiv.org/abs/2107.06751 [bsky, 2 points, 0 comments]
- A good observation, this might even be how it was generated, it's not the usual thesaurus type method of traditional tortured phrases: arxiv.org/abs/2107.06751 [bsky, 1 points, 1 comments]
- Okay so I found this article about paper mills: www.sciencedirect.com/science/arti... that doesn't talk about specific word substitutions, but it cites this paper about "tortured phrases" that does, [bsky, 1 points, 2 comments]
- tortured phrases are usually thought to be result of attempt to avoid detection by plagiarism check - they’ve been around a while and it’s likely some form of automation in the process. This is worth [bsky, 1 points, 0 comments]
- Heh. If you think this is bad, let me introduce you to tortured phrases. arxiv.org/abs/2107.06751 There used to be (and may still be) entire dedicated journals full of B.S papers that use phrases li [bsky, 0 points, 0 comments]
- Only the beginning - and on opinion would be that as part of the “game” we are also responsible arxiv.org/abs/2107.06751 [bsky, 0 points, 0 comments]
- #Fakescience arxiv.org/abs/2107.06751 [bsky, 0 points, 0 comments]
- # arXiv Tortured phrases: A dubious writing style emerging in science. Evidence of critical issues affecting established journals https:// arxiv.org/abs/2107.06751 Detection of tortured phrases in sci [mastodon, 0 points, 1 comments]
Related