Reflections on Organising Virtual SPLASH 2020
In November 2020, I was the Virtualisation Co-Chair of SPLASH 2020 (https://2020.splashcon.org/) that had a number of important innovations that I would like to share here so that future SIGPLAN conference and others can reflect upon and hopefully follow suit.
Firstly, I'd like to acknowledge that the SPLASH 2020 was a massive effort on the behalf of Hridesh Rajan as GC, Jan Vitek, Steve Blackburn, and myself as Virtualisation Co-Chairs, and very importantly Benjamin Chung (Northeastern), Wenyu Zhao (ANU), and Zixian Cai (ANU) as Video Co-Chairs. Note that on purpose we had Hridesh and Benjamin in the US, Jan Vitek in the EU, Steve, Wenyu, Zixian, and myself in Australasia to make the "timezone independent" SPLASH 2020 model work. We had a great support from Ed Nutting from CLOWDR team and Elmer van Chastelet from Conf.Researchr team who both provided their services to help SPLASH 2020 experiment with a lot of cool innovations.
When we started virtualising SPLASH 2020, we aimed for:
- Simplicity of Schedule (we didn't want the complexity that ICSE 2020 (https://2020.icse-conferences.org/program/program-icse-2020) program entailed);
- Attendance *at any time* and *from anywhere* (we didn't want the times when the confernce was "not available" that ICFP 2020 (https://icfp20.sigplan.org/program/program-icfp-2020) had despite using a form of mirroring;
- Accessibility of all talks not just by making the time work for attendees (while minimising disruption for the authors) but also by ensuring that all the talks used subtitles despite having around 250 presenters.
Here are some details of how we acheived this while trying to keep this short. An overview video of SPLASH 2020 that I recorded is available here:
And an overview of why we used specifically "12 Hour Mirror" that Steve Blackburn recorded is available here:
Simplicity of Schedule
We decided early on to give all paper talks at least 15 minutes of "talk time" with additional Q\&A scheduled afterwards because we didn't feel that it is possible to give a talk with sufficient depth of content about a technical paper if it is shorter than 15 minutes. Thus shorter talks would be a waste of both speaker and audience times unless everyone has read the paper (which is unlikely). We needed the content communicated succinctly for the Q\&A to take place afterwards.
SPLASH 2020 is a complex beast and so we made a decision to simplify dramatically by fitting the SPLASH schedule into just 3 parallel streams: "OOPSLA and ECOOP papers", "Rebase and Industry talks", "Other Major SPLASH Events" (such as GPCE, SLE, Onward! Papers, Onward! Essays, DLS, and SAS). We asked the workshops to self-organise and advertised them to the attendees but didn't ask the workshops to fit into the "pre-recroded, subtitled, mirrored" format. As a result, at our request, Conf.Researchr developed a "three column view" that we used for our program:
Attendance Anytime Anywhere
The essential motivation behind our approach is that instead of prioritising one time zone (because "most attendees are from there") we did NOT prioritise ANY time zone because if you do that then you will continue to exclude people who are NOT from your presumed "catchment area of the globe". For example, for most confernces in PL, Asia-Pacific continues to remain the "lost child" as people justify excluding them by the fact that not many people from there attend while making it hard or impossible for them to attend at the same time!
This means we really made a 100% 24 hour schedule with 2 instances of every talk separated by exactly 12 hours in such a way that ANYONE in ANY TIMEZONE can join in without sacrificing any quality during "12 hour period" of their choice. Notice that we cannot explain this enough that we did NOT have 12 hours of talks/content, we only had 8 hours of talks/content in SPLASH 2020 schedule grouped into blocks of "2 hours" with "1 hour and 20 minutes of talks/content" and "40 minutes of break times". This means that our schedule had only "8 hours of content" of which one didn't need to attend all 8 hours - that is less than a typical conference day usually includes.
The only people who had to then adjust would be the authors who did have to be there for Q\&A twice: 12 hours apart. For OOPSLA 2020, Dave Grove did a brilliant job by understanding our requirements and surveying all the authors for their "unavailable/blackout times" (e.g. from midnight to 6am in their time zone) and ensured that no OOPSLA papers were scheduled where the authors won't be comfortable getting up (say at 3am) to do a Q\&A. Unfortunately, the other events (such as ECOOP 2020) didn't quite get our message and we did end up with asking some authors to present in the middle of their night - however, this is NOT REQUIRED in this model.
We observed that in prior confernces a lot of authors whose first language is not English (e.g. a lot from Asia) much preferred to pre-record their videos so that they can practice their talk and not have to worry about being unclear. More importantly, there was a lot of mention of how they appreciated SUBTITLES when watching other people's talks as it is hard to deal with accents and slang employed by other speakers. We also noted the advances in FREE technology for automatic subtitle generation such as YouTube Studio that meant that we simply asked the authors to pre-record all their talks, then upload and process them for free with YouTube Studio (or other free platforms) and then hand edit their own subtitles and provide us with both video (MP4) and subtitle (SRT) files!
We then put together a continuous video stream with talks fitted into the exact slots (starting at the exact :00, :20, and :40 times) with filler slides as required and streamed the first 5 minutes of Q\&A following each talk in a stitched together continuous stream. We need to observe here that using essential tools like Clowdr and experience in video broadcasting using tools like OBS is essential here as well as having people across the world available 24/7 in the time comfortable for them to make this work.
I feel really strongly that all the other conferences should follow this example on accessibility and ensuring that we do not exclude people in "inconvenient time zones" (such as Asia-Pacific as the case seems to be) as we have a HUGE number of PL researchers doing good work in China, Japan, Korea, Singapore, and many other countries in this part of the world (including Australia and New Zealand (where I am from)). I think that trying to run a conference at a time that is not "spread around" like SPLASH 2020 did but rather trying to find "perfect spots" or "small number of hours" that accommodates "a lot of people" will always exclude some people and is thus not a fair solution to the "globally inclusive world" that we can build using the virtual conferences movement. We should be able to demonstrate with our efforts that virtual conferences can in fact be run better than in person conferences both increasing inclusiveness and global connections without significant carbon footprint or travel requirement. But we need to stop acting as a gap filler "during Covid times" expecting everything to go back to 2019 "normality" and rather use this opportunity to make choices that are expected from all organisers and authors/presenters to have conferences like SPLASH (which we capped at 1000 attendees for technological reasons but could easily be in the 1000's) run in such a way that we have large numbers taking part without the barriers of TIME and COST. Of note, SPLASH 2020 had India and Japan as top participants close behind US and EU according to our video logs - while in real life SPLASH, we rarely get many people coming from Asia-Pacific.
Finally, as someone who is looking after two children and having to take them to all the sports and events after school myself as my partner has a less flexible job than a Professor, I feel really strongly about making it possible for people to attend without having them staying up all night while looking after their children during the day. As a result, Jonathan Aldrich, Steve Blackburn and I started a petition today calling for ACM to change how it runs virtual CS conferences starting with ACM SIGPLAN asking POPL, PLDI, ICFP, and SPLASH to accommodate all time zones equally around the world. Please see our petition here:
and please sign it using this form:
Post a Comment