The expression “tortured phrases” likely reminds teachers of student essays. Paradoxically, it takes time to develop a “natural” style. But last April, when that sort of language started appearing in computer science journals, some alert researchers suspected that something more serious than mere awkwardness was at work:

The researchers could not understand why researchers would use the terms ‘counterfeit consciousness’, ‘profound neural organization’ and ‘colossal information’ in place of the more widely recognized terms ‘artificial intelligence’, ‘deep neural network’ and ‘big data’.

Holly Else, “‘Tortured phrases’ give away fabricated research papers” at Nature

But they figured it out. Many computer science papers, especially from China, were partly constructed using automated translation and software that may disguise plagiarism.

But the software can’t tell the user whether the phrase would sound odd to an English-speaking colleague. For example, “remaining energy” became “leftover vitality”; “cloud computing” became “haze figuring”; and “signal to noise” became “flag to commotion.” Such off-beat phrases in place of conventional ones — while not wholly uninformative — were bound to attract attention…

Guillaume Cabanac of the University of Toulouse told Nature that his team found over 860 publications that included at least one tortured phrase, of which 31 were published in Microprocessors and Microsystems. Publisher Elsevier has been investigating and agrees that reverse translation, intended to disguise plagiarism, was probably the source of the unusual expressions.

The suspect phrases have turned up in hundreds of journals. Checking them all out wouldn’t be easy. Research integrity analysts have called the find “shocking” and “the tip of the iceberg.”

The same AI that enables much easier communications also enables much easier intellectual theft. And there is no simple solution in sight, especially because automated translation may become more sophisticated, enabling plagiarism that is less easy to detect.

The paper is open access.

You may also wish to read: New AI can create—and detect—fake news. But how good is it at either task? We tested some copy. Will the predicted tsunami of fake news and advertising make much difference? Possibly, but in ways that might surprise you. (August 7, 2019)