“Tortured Phrases” give away fabricated research papers

0


[ad_1]

In April 2021, a series of strange phrases in magazine articles piqued the interest of a group of computer scientists. The group, led by Guillaume Cabanac from the University of Toulouse in France, could not understand why researchers use the terms “fake awareness”, “profound neural organization” and “colossal information” instead of the generally accepted terms “artificial intelligence” ‘,’ Deep Neural Network ‘and’ Big Data ‘.

Further research revealed that these strange terms – which they call “tortured phrases” – are likely the result of automated translation or software that tries to cover up plagiarism. And they seem to be widely used in computer science papers.

Scientific Integrity Detectives say Cabanac and his colleagues discovered a new type of fabricated research and that their work was preprinted on July 12th on arXiv1, could just be the tip of the iceberg in terms of the literature concerned.

To get a feel for how many articles are affected, the researchers searched for 30 tortured phrases in magazine articles indexed in the Dimensions citation database. They found more than 860 publications that contained at least one of the sentences, 500 of which were published in a single journal: Microprocessors and microsystems.

“That harms science. You can’t trust these papers, so we have to find them and move them in, ”says Cabanac.

Suspecting that the tormented phrases are the result of an automated translation or software that rewrites existing text, Cabanac and colleagues have a selection of abstracts from Microprocessors and microsystems and other magazines via a tool that can detect whether texts have been generated using the artificial intelligence tool GPT. Of Microprocessors and microsystems Papers marked by the tool resulted in manual checks in some of them “critical errors” such as nonsensical text as well as copied texts and images.

To dig deeper, the group downloaded all of the articles published in Microprocessors and microsystems between 2018 and 2021, a period they chose because an updated version of GPT was released in 2019. The analysis found that articles published after February 2021 had an average adoption time five times shorter than those published before that date. A large proportion of this work comes from authors from China. And a subset of papers had identical submission, revision, and acceptance dates, most of which appeared in special editions of the magazine. That is suspicious, say the authors. In contrast to standard editions, which are supervised by the editor-in-chief, special editions are usually proposed and supervised by a guest editor and focus on a specific area of ​​research.

Microprocessors and microsystems wasn’t the only title affected – the researchers also found evidence of tortured phrases in articles published in 35 other journals. “Preliminary research shows that several thousands of articles with tortured phrases are indexed in large databases,” they write, adding that “other tortured phrases relating to the concepts of other scientific fields have yet to be uncovered.”

Special investigation

Around the time Cabanac and his colleagues first noticed the tormented phrases, and without their knowing it, became the editor of Microprocessors and microsystems began to have concerns about the integrity and rigor of peer-reviewed articles published in some of the special editions of the journal.

The magazine’s editor, Elsevier, opened an investigation. This is still ongoing, but in mid-July the publisher added its concern to its 400+ articles that have appeared in six special editions of the magazine.

The concerned statements state that the papers in the affected special issues of the Microprocessors and microsystems will be progressively “independently re-evaluated” and the journal will publish further updates on its status after the research is complete.

A “configuration error in the editorial system” of the magazine had meant that neither the editor-in-chief nor the editor in charge of editing received it for approval as intended, the publisher said. “This configuration bug was a temporary system migration issue and was fixed as soon as it was discovered,” the note said.

An Elsevier spokesman said nature in a statement that the Microprocessors and microsystems Research has found that the authors likely used back translation software to cover up plagiarism, and that this is the likely source of the tormented phrases.

The investigation also found that 49 articles flagged as suspicious by Cabanac and his colleagues and published in the magazine’s standard editions were originally submitted in special editions and accepted by guest editors, “but were then published in regular editions by the authors.” ‘Inquiry,’ the statement said. These papers are already part of Elsevier’s investigation, she adds.

Elisabeth Bik, a California Scientific Integrity Analyst who is known for her ability to spot duplicate images in papers, says the results of Cabanac’s research are “shocking.” “This is a very new and unsettling kind of manufactured paper,” she adds.

Jennifer Byrne, a molecular oncology researcher at the University of Sydney, Australia who is also working on the discovery of fabricated papers, says this is likely the tip of the iceberg because the researchers only looked closely at one publisher’s journal. “These papers were also found because they were of very poor quality, but there could be more plausible AI-generated papers in the literature that are harder to see,” she adds.

[ad_2]

Leave A Reply

Your email address will not be published.