New signal detects tortured phrases in manuscript submissions

Signals now scans journal submissions for tortured phrases, helping editors and research integrity teams quickly spot these issues early in the publishing process. These unusual and often nonsensical phrases are used in place of standard academic terms — a couple of examples include “man-made consciousness” instead of “artificial intelligence”, and “bosom peril” instead of “breast cancer.” Tortured phrases can be a clear indicator of publishing misconduct, often used to evade plagiarism detectors.

Spotting tortured phrases is an ongoing challenge for editorial and research integrity teams — thousands of articles published in 2024 contain them. While the phrases are clearly problematic once identified, they are easily missed by individuals reading through long articles under time pressure. 

For high accuracy, this task requires an automated approach, which is available as part of Signals Manuscript Checks. Signals compares manuscript text against the frequently updated and well-established list of tortured phrases compiled by Guillaume Cabanac, who runs the Problematic Paper Screener with his team. Signals flags how many tortured phrases are in a manuscript and lists them, making the process quick and simple for editorial teams.

Thanks to data from the Problematic Paper Screener, Signals has flagged tortured phrases in published articles for the past year. This has proven to be a valuable signal when we help publishers understand integrity risks in their portfolios. Detection of tortured phrases in Signals Manuscript Checks has been a frequent request, and we’re pleased to make it available as part of our ongoing rollout of features that evaluate manuscript content — including the risky email signal and Sleuth AI — which help editorial teams make quick decisions that protect journal reputations and the scholarly record.

The tortured phrases signal is now available to all publishers using Signals Manuscript Checks, via both ScholarOne integration and Direct Upload.