Netizens condemned ACL chairman for publicly criticizing arXiv and suggested considering other top conferences

Emily M. Bender, a professor at the University of Washington and Vice Chairman of the ACL 2023 Conference, caused a huge controversy with her remarks a few days ago

arXiv has ruined the research atmosphere, then What are we brushing every day? Boaz Barak, a professor at Harvard University, said that arXiv has done much more to promote scientific progress and expand scientific participation than all anonymous interventions. Any policy that hinders arXiv is not only stupid, but also undermines scientific progress and inclusion. Sex has the opposite effect

Deep learning giant Yann LeCun supported his views

The discussion about Emily Bender’s views is actually not completely unreasonable . After all, arXiv is a platform with almost no review: if you submit a manuscript to an academic conference or journal, the author will remain anonymous during the review period, but on arXiv people always involuntarily follow big-name research institutions and scholars. However, since what happened yesterday, the direction of the discussion has completely shifted to one side

In order to understand the ins and outs of the matter, we need to first take a look at the ACL paper that caused an uproar in the AI ​​community yesterday. Rejection incident

Naomi Saphra, a postdoctoral fellow at New York University (studying with Kyunghyun Cho) posted on X on September 4, saying that her paper had just been rejected by ACL, the top international conference in the field of natural language processing. This was her first conference paper submission with students, and she had previously told the first author that top conferences on natural language processing were more user-friendly than machine learning conferences. During the review period, the first author spent a lot of time to refute, and the experimental results also supported the point of view of the paper

However, the most important thing is that the reason why she was rejected was because she was under the anonymity policy The paper was submitted to arXiv after the deadline

Naomi Saphra said that the ACL policy has caused harm to new researchers and caused the conference to deviate from the original intention of the field of natural language processing

There has been a long-standing controversy over AI’s ban on publicizing papers during review. Nowadays, in the AI ​​circle, many people are accustomed to staring at the preprint paper platform arXiv every day, because this platform is updated every day. In contrast, research releases at the annual top conference are much slower

In the field of AI, if you don’t read the news, you will fall behind

After Naomi Saphra complained, people began to criticize the current order The annoying AI conference policy

Sebastian Raschka, a well-known machine learning researcher and assistant professor at the University of Wisconsin-Madison, conducted some research. He pointed out that the rejection of this paper was not due to missing the conference paper submission. The deadline was a few minutes, but because it was submitted a few minutes late to an optional third-party preprint paper platform. What's going on?

Previously they didn't allow people to discuss their arxiv version of the paper online, now can they also limit when people can upload to arxiv?

Sebastian Raschka found the review policy specifically as follows: At ACL 2023, papers submitted through direct methods have an anonymity period and are not allowed to be published in preprints Signatures are made public on the paper platform. The times specified on arXiv refer to the time of submission, not the time of publication on arXiv. The review policy discovered by Sebastian Raschka is specifically this: at ACL 2023, papers submitted through direct submission will have an anonymity period and will not be allowed to be publicly signed on the preprint paper platform. The designated time refers to the time of submission on arXiv, not the public time on arXiv

Gautam Kamath, a professor at the University of Gaotielu, also said, I think this policy is very stupid, and In actual implementation, the situation was worse than imagined

Now that he has set a deadline for uploading to arXiv, papers by Naomi Saphra and others were rejected because they missed this time. You see, he has now set a deadline for you to upload to arXiv. It is because of missing this time that the paper of Naomi Saphra and others was rejected

Boaz Barak sarcastically asked, is it possible that the author of the paper needs to go to the monastery to stay before the review results come out?

We are currently in the era of large models, and competition in the NLP field has become extremely fierce. The careers of many young researchers depend on displaying their work on high-visibility platforms in the hope of getting more citations. However, the current practices of top artificial intelligence conferences are contrary to this

According to the original intention of conferences such as ACL, it is obviously to promote academic exchanges and promote the development of advanced research. However, the painful policy has even made people question the authority of the conference itself

Google VP Fernando Pereira said: ACL has long ceased to be academic, and now it is a bureaucracy. route.

PhD student Fu Yao said, I participated in two conferences, ACL and ICML, this year, and there is a sharp contrast between the two. The content of ACL 2023 is basically more than a year out of date, while the content of ICML is full of prospects for the future. If the younger generation of natural language processing researchers wants to have a future, they should submit their articles to NeurIPS/ICLR/ICML

Someone mentioned that the regulations that led to the rejection of Saphra's paper date back to 2017 . Compared with the rapid development of AI technology, the development of rules for AI conferences seems to be relatively slow.

Interestingly, ACL has not actually changed very slowly. Since 2022, ACL has launched a mechanism called "rolling review", namely ACL Rolling Review (ARR), which is similar to the review process of journals. Recently, ACL has revised its regulations. The 2024 conference will completely cancel direct submission. Everyone can only choose to submit ARR papers.

Previously, papers submitted directly to the ACL conference It can be processed according to the regular review process: a paper will be reviewed by 3 or more reviewers, and the author can submit a reply to respond to the review comments. Once the paper is accepted, the author can edit the final draft

Under the ARR mechanism, submitted papers will be processed by the Senior Area Chair (SAC). The author can provide a reply, but cannot modify the paper. The paper needs to obtain all review comments and comprehensive review comments (meta-review) and submit it to ACL before the deadline. The content cannot be modified, but the author's reply can be attached. Once submitted to ACL, it cannot be modified and submitted to ARR again (unless the conference gives an acceptance result)

According to the current trend, ACL will continue to develop in the direction of multiple rounds of reviews in the future to be more Close to the standards of academic journals

Currently we have not found a solution to balance high-quality research and rigorous and fair review speed. Although rolling review has tried its best, it still cannot compare with the speed of arXiv

The following is the reference content:

Please click the following link to view detailed information about the 2023 ACL Main Conference: https://2023.aclweb.org/calls/main_conference/

The above is the detailed content of Netizens condemned ACL chairman for publicly criticizing arXiv and suggested considering other top conferences. For more information, please follow other related articles on the PHP Chinese website!