Hundreds of Gibberish Papers Still Lurk in the Scientific
Total Page：16
File Type：pdf, Size：1020Kb
News in focus says Coudert. He had previously posted about says she was surprised and angry to learn that of Toulouse, France, came up with a new idea: the Nature story on Twitter; McNutt replied, scientists continued to collaborate with the searching for key grammatical phrases char- urging him to take action. astronomer, pointing out that manuscripts acteristic of SCIgen’s output. Last May, he and “The NAS has chosen a policy that is very posted on the arXiv preprint server in the past Cabanac searched for such phrases in millions weak and that protects them in a way,” says six months still listed Marcy as a co-author. of papers indexed in the Dimensions database. Coudert. “Where is the justice for women pushed out of After manually inspecting every hit, the The academy has said in the past that it does the field if people continue to work with him?” researchers identified 243 nonsense arti- not have the resources for formal investiga- Some of those papers point to Berkeley as cles created entirely or partly by SCIgen, tions, apart from for internal NAS business. Marcy’s affiliation. A Berkeley spokesperson they report in a study published on 26 May The group relies on publicly documented (G. Cabanac and C. Labbé J. Assoc. Inf. Sci. investigations carried out by other organiza- “Where is the justice for Technol. https://doi.org/gj7b8h; 2021). These tions to begin inquiries into its members. articles, published between 2008 and 2020, The NAS informed Coudert of Marcy’s ter- women pushed out of the appeared in various journals, conference pro- mination last month; the chemist says it is a field if people continue ceedings and preprint sites, and were mostly in preliminary step in the right direction. to work with him?” the computer-science field. Forty-six of them Membership of the highly selective NAS had already been retracted or deleted from the is regarded as a top honour in US science, websites on which they were first published. burnishing the profile of elected members. It says that Marcy is currently a retired professor Since last year, the researchers have added also confers a degree of influence — the group at the university, and that University of Cali- another 20 papers to their list, including is regularly tapped by US agencies to offer fornia policy allows retirees to refer to them- gibberish articles created by MATHgen (soft- scientific views on national affairs. selves as emeritus faculty members at those ware that generates mathematics papers) and Seyda Ipek, a theoretical particle physicist institutions. They added that the university’s the SBIR proposal generator (which creates at the University of California, Irvine, also sub- 2015 announcement of Marcy’s resignation nonsense grant proposals). mitted a complaint last September, including was accurate at that time. public details of harassment investigations A spokesperson for the NAS confirmed that CV padding and findings concerning Marcy. “It’s really Marcy’s membership had been rescinded as Most of the latest batch of SCIgen papers were important to not allow these people in these of 24 May. They did not say how many other authored by researchers from China (64%) or prestigious communities, because they are members were under review as a result of India (22%), although Labbé notes that the doing bad things for science,” says Ipek. She sexual-harassment complaints. manuscripts could have been submitted in anyone’s name without their knowledge. One author of several of the papers told Labbé and Cabanac that he’d submitted them as hoaxes. But other manuscripts seem to have been edited with genuine reference lists, suggest- HUNDREDS OF GIBBERISH ing that they might have been generated to inflate scientists’ citation counts. “I think the PAPERS STILL LURK IN THE vast majority are created to pad CVs in order to fulfil a need to publish papers,” says Labbé. SCIENTIFIC LITERATURE The researchers found only two SCIgen papers that hadn’t been retracted at the IEEE Nonsensical articles, spotted years after the problem — which is evaluating both of them — and one Springer paper that included a fragment of was first seen, could lead to a wave of retractions. MATHgen text. But other publishers were caught out more badly. IOP Publishing, a By Richard Van Noorden proceedings published by the Institute of subsidiary of the London-based Institute of Electrical and Electronic Engineers (IEEE); Physics, says it retracted ten papers “as there onsensical research papers gener- he went on to find more than 120 fake SCIgen was clear evidence they had been comput- ated by a computer program are still papers published by the IEEE and by Springer er-generated”, and is investigating why they popping up in the scientific literature (C. Labbé and D. Labbé Scientometrics 94, 379– weren’t identified during peer review at the many years after the problem was first 396; 2013). It was unclear who had generated conference for which they were accepted. “We seen, a study has revealed. Some pub- the papers or why. The articles were subse- have reasonable evidence to suggest that the Nlishers have told Nature they will take down quently retracted — or sometimes deleted — peer-review process for some of these papers the papers, which could result in more than and Labbé released a website allowing anyone was compromised,” says Kim Eggleton, the 200 retractions. to upload a manuscript and check whether it publisher’s integrity and inclusion manager. The issue began in 2005, when three PhD seems to be a SCIgen invention. Springer also The publishers that posted the most SCIgen students created paper-generating software sponsored a PhD project to help spot SCIgen content were Trans Tech Publications based in called SCIgen for “maximum amusement”, and papers, which resulted in free software called Bäch, Switzerland, which published 57 SCIgen to show that some conferences would accept SciDetect. (Springer is now part of Springer papers; Blue Eyes Intelligence Engineering and meaningless papers. The program cobbles Nature; Nature’s news team is editorially inde- Sciences Publication (BEIESP), based in Bho- together words to generate research articles pendent of its publisher.) pal, India, which had 54; and Atlantis Press, a with random titles, text and charts, easily spot- Labbé, who works at the University of Paris-based publisher that was acquired by ted as gibberish by a human reader. It is free to Grenoble Alpes in France, originally searched Springer Nature this March, with 39. Both Trans download, and anyone can use it. manuscripts for words typical of SCIgen’s Tech Publications and Atlantis told Nature that By 2012, computer scientist Cyril Labbé had vocabulary. But he and another computer they were investigating and were in the process found 85 fake SCIgen papers in conference scientist, Guillaume Cabanac at the University of retracting the articles, but a spokesperson 160 | Nature | Vol 594 | 10 June 2021 ©2021 Spri nger Nature Li mited. All ri ghts reserved. ©2021 Spri nger Nature Li mited. All ri ghts reserved. for BEIESP said that it published only articles SCIgen papers are extremely rare: Labbé and 600 million doses have been delivered. Sino- with original content that passed double-blind Cabanac estimate from their screen that they pharm’s vaccine has been approved in many peer review and plagiarism checks. make up a mere 75 papers per million in the more nations. But WHO emergency approval The SSRN preprint server, where papers computer-science literature. But, says Labbé, could now facilitate the further distribution are shared before peer review, had published the existence of these papers is an indication of both vaccines to low-income countries, 16 SCIgen articles, the study found. A spokes- of the harmful effects of a ‘publish or perish’ through the COVID-19 Vaccines Global Access person for SSRN said it was investigating the culture, and an example of how nonsensical (COVAX) initiative. issue, and noted that it provided “limited work can still make it into conference pro- A spokesperson for COVAX member Gavi, the screening” for its preprints (with “advanced ceedings or journals. “You shouldn’t find these Vaccine Alliance, in Geneva, Switzerland, says screening” for health-care manuscripts). things in the literature,” he says. Gavi welcomes the news of the listing, “as this means the world has yet another safe and effec- tive tool in the fight against this pandemic”. The WHO’s efficacy estimate of 51% was based on data from late-stage trials among health-care workers in Brazil, posted online CHINA’S CORONAVAC JAB as a preprint in April (R. Palacios et al. Pre- print at SSRN https://doi.org/ggjr; 2021). Of SET TO BOOST GLOBAL the 9,823 participants included in the analysis, 253 had COVID-19 — 85 in the vaccinated group IMMUNIZATION CAMPAIGN and 168 among those who received the pla- cebo. None of the vaccinated volunteers World Health Organization approves second of two was hospitalized or died owing to COVID-19. Smaller, late-stage trials in Indonesia and Tur- Chinese COVID shots in use in more than 70 nations. key have shown higher efficacies, of up to 84%. Preliminary findings from a post-trial study By Smriti Mallapaty the global fight against COVID-19 as a safe and of 2.5 million people in Chile estimated that moderately effective SARS-CoV-2 vaccine,” says CoronaVac was 67% effective at preventing he World Health Organization (WHO) has Murat Akova, a clinical researcher in infectious COVID-19, and 80% effective at preventing approved a second Chinese vaccine for diseases at Hacettepe University in Ankara. death from the disease, despite the presence emergency use. CoronaVac was found to CoronaVac’s approval, on 1 June, came of the Alpha (B.1.1.7) and Gamma (P.1) variants be 51% effective at preventing COVID-19 about a month after the WHO listed another of the virus SARS-CoV-2.