Minutes from NAACL Exec Meeting at NAACL 2024
- Date: Wednesday June 19, 2024
- Time: 11:00am - 12.00pm Central Standard Time (UTC-6)
Participants
- Graham Neubig
- Luciana Benotti
- Greg Durrett
- Jonathan May
- Jessy Li
- Yunyao Li
- David Yarowsky
Position with respect to ML conferences
- Who is joining each type of conference?
- Many of the people publishing on LLMs never submitted in *ACL conferences. They are used to NeurIPS or ICML. But some people who traditionally published in both ML and NLP do not go to NLP conferences anymore. Do we want to attract those people back?
- On the other hand, there are more people from HCI coming to NLP conferences, more people are working on human eval problems. We could have a special track to motivate people to come back.
- Why do big companies not have booths here?
- Barriers to publishing in/attending NAACL
- The Anonymity policy was a big barrier for people to publish but this is not relevant anymore.
- ARR is a barrier to access for unfamiliar people due to its complexity.
- The tracks in the CFP do not reflect recent research trends such as LLMs.
- There is a perception that more people pay attention to papers in ML conferences.
- There is a tradeoff in location: locations in the US encourage participation from US industry, but discourage participation from, e.g. Latin America.
- Industry submissions (eg internship papers) sometimes do not work well with *ACL deadlines in their current state but work with ICLR.
- A regional meeting is perceived as second tier.
- Possible mitigation strategies to encourage more participation
- A special track on topics such as alignment.
- Invite somebody from the ML community as a keynote speaker.
- Fix the tracks to consider LLMs more comprehensively.
- Orals should not be randomly selected. So orals should be the best papers. In NeurIPS ACs decide what is the best format to present the paper.
- Two deadlines for conferences might be good.
- Best paper awards should be announced earlier and have a special oral talk.
- We can reach the COLM community and they could potentially collocate with an ACL conference.
- As part of renaming the chapter we could also rename the conference, it does not need to be the same name as the chapter.
- We could use some good aspects of ARR, openreview and a rolling review pool, but have direct submission without a commitment.
- Have a closed-door session between industry and academia, interactive panel where industry understands how to work more productively with academia.
- Possible strategies to attract sponsors
- Have organizers from potential sponsors, such as people in industry.
- Find sponsorship chairs that are well connected to potential sponsors.
- Activate the NAACL sponsorship chairs that have a 2-year term so that they find sponsorship for NAACL.
- Our reviewers and authors will go to ML conferences such as ICLR. The most important thing is to get a critical mass of people joining the conference.
Virtual attendance
- People are often too tired/busy to attend both an in-person and virtual conference, it might be better to combine both.
- We can have virtual posters that are on a screen, but in previous iterations these were in a different place than the normal posters. The problem is the noise and they have to be separated.
- We could invite people from industry to attend our virtual parts of the conference.