ICLR and ARR limit the number of papers per author — Science News

The International Conference on Learning Representations (ICLR) and the ACL Rolling Review (ARR) system have limited the number of scientific papers a single researcher may submit for review. The decision was made amid a sharp increase in the number of papers and a shortage of reviewers, Science News reports.

ICLR, one of the largest conferences in artificial intelligence, received about 500 submissions 10 years ago. After the submission deadline for the 2027 conference, the number of submissions exceeded 40,000. Organizers attribute the surge to growing interest in AI, pressure to publish, and the ability to prepare texts more quickly with the help of artificial intelligence.

Limits for authors and reviewers

ICLR and ARR have set a limit of 20 papers per submission cycle for most authors. The ICLR rule will take effect for the 2027 conference. A researcher will also be able to author only one paper whose co-authors do not include a qualified reviewer.

ARR, which conducts reviews for the annual Association for Computational Linguistics conference and related events, is introducing new rules this month. They limit the number of submissions on which a person may be the first author to five. If none of the authors meets the reviewer criteria, they may involve a qualified colleague, but one person may serve in that role for no more than two papers.

More current news is available on the UA.News Telegram channel Telegram.

Shortage of reviewers

This year, 20% of the 19,500 submissions to ICLR had no author who could perform reciprocal reviewing. Many of these researchers were at the beginning of their careers and had no publication record. In May, about 38% of the 17,000 papers submitted to ARR also had no qualified reviewer among their authors.

ARR considers a reviewer qualified if they have previously published two papers at AI conferences from a specified list. ARR had to ask such participants to take on more reviews than would be reasonable. Papers with no author meeting the established criteria are directed by the system to a lottery, from which only some will be selected for expert evaluation.

The limits are intended to encourage authors to submit stronger papers, but they may make participation more difficult for early-career researchers from large research groups who co-author many papers. ICLR organizers also acknowledge the risk of nominal co-authorship by qualified reviewers who did not participate in the research and said they would monitor this.

Other platforms are taking similar steps. Last year, NeurIPS began requiring authors to complete their assigned reviews before they could view feedback on their own papers. Last week, the arXiv preprint server announced a limit of two papers per author per month: in September, it received 40,363 manuscripts, compared with 20,569 in September 2024.