Tortured phrases in scientific publications refer to unexpected and unusual phrases that are used in place of established scientific terms, often as a result of automated translation or software that attempts to disguise plagiarism. These phrases have been identified as potential indicators of fabricated research papers. For example, “counterfeit consciousness” may be used instead of “artificial intelligence,” and “colossal information” instead of “big data”[1]. Guillaume Cabanac, a computer scientist at the University of Toulouse, France, has conducted research on this phenomenon and has uncovered over 3,000 papers containing tortured phrases, some of which were published in reputable journals. He and his colleagues have developed tools to detect and uncover such problematic science, emphasizing the need to raise awareness and take steps to decontaminate the scientific literature[3].
The use of tortured phrases is associated with deception, misconduct, and fraud in science. Researchers have found that some authors may use advanced language models and machine-paraphrasing tools to produce papers with tortured phrases in an attempt to bypass regular anti-plagiarism checks. To detect these unreliable and often fake papers, researchers have developed tools such as the “Problematic Paper Screener,” which scrutinizes a large number of publications to uncover papers with tortured expressions[3].
The presence of tortured phrases in scientific publications poses a threat to the integrity of the scientific literature. It is important to detect and address these problematic papers to maintain trust and credibility in academic publishing. The responsibility to combat this issue falls on both those who evaluate researchers and the researchers themselves. While efforts are being made to detect and uncover problematic science, the use of large language models to generate text could further complicate the problem, making it essential to grow a community dedicated to detecting problematic science and raising awareness about how to address it[3].
More Examples of Tortured Phrases
Here are additional examples found in AI written scientific journal articles.
- Arbitrary woods for random forests
- Bleakness for morbidity
- Blubbery acids for fatty acids
- Blunder rate for error rate
- Bosom peril for breast cancer
- Burglar nobles for robber barons
- Colossal information for big data
- Corridor impact sensor for Hall effect sensor
- Counterfeit consciousness for artificial intelligence
- Discourse acknowledgement for voice recognition
- Flag to clamor for signal to noise
- Fluffy rationale for fuzzy logic
- Haze Figuring for cloud computing
- Irregular esteem for Random value
- Kidney disappointment for kidney failure
- Leftover vitality for Remaining energy
- Lymph hub for lymph node
- Mean square blunder for mean square error
- Motion to clamor for signal to noise
- Motor vitality for kinetic energy
- Numerically interesting for statistically significant
- Ovarian disease for ovarian cancer
- Profound neural organization for deep neural network
- Random degeneration for accidental degeneracy
- Shrewd devices for smart devices
- Social separating for social distancing
- Spewing for vomiting
- Subterranean insect province for ant colony
- Sun-orient force for solar energy
- Tainted people for infected individuals
- Top notch picture for high-resolution image
- X-beam for X-ray
Citations:
[1] https://languagelog.ldc.upenn.edu/nll/?p=51739
[2] https://arxiv.org/abs/2107.06751
[3] https://www.dimensions.ai/blog/detecting-tortured-phrases-to-unmask-fake-science/
[4] https://arxiv.org/pdf/2107.06751.pdf
[5] https://www.editage.com/insights/tortured-phrases-what-they-are-how-they-are-detected-and-how-to-avoid-them