2020Q3 Reports: Publicity Chairs

ACL 2020 post-conference report: Publicity Co-Chairs

Emily M. Bender, Esther Seyffarth, Zhiyuan Liu

The publicity team for ACL 2020 was set up so that the three members covered different time zones, which was especially useful for the virtual conference format. The team's activities focused on the following areas:

Pre-conference publicity

The initial phases of the publicity chair role involved disseminating information about the conference to potential authors, reviewers, and attendees well ahead of the event. This was done via email (especially the ACL Portal and the corpora mailing list, but also a collection of other mailing lists) and social media (primarily Twitter and also Facebook). The emails were sent from the announcements@aclweb.org email address. Emily has maintained a spreadsheet of which mailing lists/web services were contacted when for major announcements (e.g. call for papers), which can be shared with the next publicity chair(s). Several of the mailing lists require subscription in order to be able to post, and so the inbox of this email address probably needs cleaning periodically.

Beginning with ACL 2020, and under the leadership of the ACL Publicity Director Barbara Plank, we have moved to creating stable social media accounts that will persist from year to year, avoiding the need to build up a following each time around. As of July 9, 2020, the @aclmeeting account on Twitter has 6,333 followers and the Facebook page has 349. The newly created Weibo account (created just before the conference started) has 1,638 followers. We did not make use of Instagram this year, having initially planned to save that for photo content at the conference and then finding this moot given the on-line format.

Twitter

Emily M. Bender and Esther Seyffarth were the main posters on the official ACL Twitter account before and during the conference. We posted announcements coming from the PCs, reminders for conference events, and tips about participating in the virtual conference.

We also organized a team of live microblogging volunteers with the goal of making research presented at the conference visible to a wider community. The microblogging efforts were inspired by previous conferences (especially NAACL 2019), and some aspects were added/changed:

1. Since talks were pre-recorded, microbloggers were able to watch and summarize them at their own convenience. We encouraged microbloggers to do so before the Q&A associated with the paper, if possible.
2. Microbloggers were invited to post in other languages than English, which led to the use of roughly 20 different language-specific hashtags. (We used the format #acl2020xx where xx was the BCP-47 language code; all languages in our sample had two-letter codes through BCP-47.) This initiative was well-received, with microbloggers reporting connections to NLP research communities they hadn’t been connected to before because of their non-English tweets. Microbloggers also expressed excitement about having this platform to reach out to their communities in this way.
3. Microbloggers were not restricted to Twitter as in past conferences, but were free to choose platforms on which to post. The platforms with the highest number of microbloggers for the conference were Twitter (32 volunteers) and Weibo (8 volunteers).
4. Microbloggers were not assigned to conference sessions (impossible in the virtual format), but instead to topic areas. They were asked to cover at least 5 papers from their area. With the help of Matt Post from the ACL Anthology, we provided a list of first-time first authors at ACL to microbloggers, and asked them to include at least one of these "newcomers" among the first three papers they covered.

We held a one-hour microblogging tutorial over Zoom in the week leading up to the conference, and provided written notes to the volunteers to guide them. We recorded this presentation to share with microbloggers who couldn’t attend live and can share it with the next publicity chair(s) should that be helpful. Information (profile URLs, names, etc.) was collected from live microbloggers using a Microsoft Office form.

During the conference, we retweeted Twitter threads about papers from the official @aclmeeting Twitter account. We used an automation service (Zapier) to watch the hashtag on Twitter and add information to a spreadsheet to help us keep track of which tweets had already been retweeted. This provided one kind of benefit to the microbloggers, in the form of increased visibility. We believe that the retweeting of the non-English tweets also raised the visibility of ACL as an international and inclusive organization. One suggestion we have for future conferences that adopt this is to establish a system to allow the “official” microbloggers to signal to the publicity chairs that specific tweets are just personal commentary about an ACL paper and shouldn’t be retweeted. (For the most part, we took the lack of inclusion of the language-specific hashtag to signal this, but when the tag for English was left off, and the microblogger was tweeting about a specific paper, it was less clear.)

Weibo

For the first time, an official ACL account on Weibo (https://weibo.com/aclmeeting) was created and used throughout the conference, with Zhiyuan Liu as the main poster on that account, Xu Han, Fanchao Qi, and Zhengyan Zhang as the assistants during the conference. We posted announcements from the PCs, reminders for conference events, and tips about participating in the virtual conference, most of which coordinate with Twitter posts. We posted overall 70 messages before and during the conference.

Press

We contacted members of the press in each of our countries and offered free conference registration to them if they were willing to cover the conference in their medium. One German journalist, three Chinese journalists, and two journalists from English-language media registered and participated in the conference.

To make the conference topics more accessible to journalists, we provided two resources:

1. A collection of lay summaries by authors of ACL 2020 papers. Each summary was up to 500 words long. The summaries were grouped by topic area.
2. A selection of papers deemed to be especially likely to hold interest for the general public, provided by the PCs.

Journalists were encouraged to use these resources and the virtual ACL2020 website to form an impression of the conference, and to contact authors for interviews if they wanted. Authors were given tips on how to interact with the media, in the form of an ACL 2020 blog post published just before the conference. We also created a #press-room channel in the RocketChat as a place for journalists and authors to meet.

The response to the invitation to write lay summaries was extremely positive, and we recommend carrying this out again in the future, not least because (as authors reported) the exercise of writing a lay summary is a useful one.

It is unclear to us how many journalists attended the conference, and the #press-room channel was very quiet. Nonetheless, ACL 2020 did appear in at least some news articles in English:

• https://syncedreview.com/2020/07/08/acl-2020-announces-best-paper-test-of-time-awards/
• https://slator.com/academia/and-the-winner-is-acl-2020-announces-best-paper-awards/
• https://venturebeat.com/2020/07/09/ai-researchers-create-testing-tool-to-find-bugs-in-nlp-from-amazon-google-and-microsoft/

and Chinese:

• https://mp.weixin.qq.com/s/8ju0eVuJOurBphly9sSPLQ
• https://mp.weixin.qq.com/s/FrrlawFBvf8eBVpuVlF3Cw
• https://mp.weixin.qq.com/s/HRANdENP04F-6eLG7VjCKg
• https://mp.weixin.qq.com/s/Cp-I4WAANW_K29uRTDaQOg
• https://mp.weixin.qq.com/s/QsdR8MncJpLXsD03585rMg

RocketChat

The activities of the publicity team on RocketChat fell mainly into three categories:

1. Making announcements in the #announcements channel, usually containing the same information as announcements posted on Twitter. In contrast to pre-conference announcements, where members of the organizing committee sent info to the publicity chairs for social media, for the in-conference RocketChat, we encouraged committee members to also make announcements directly (to reduce latency).
2. Providing a #press-room channel as a space for researchers and journalists to interact.
3. Providing a #social-media-posts channel where posts by the official microbloggers were cross-posted. The aim of this channel was to make Twitter content visible to researchers in China (for whom Twitter is blocked) as well as conference participants without social media accounts, and to allow these groups to be able to watch the conversation going on in social media. Only posts by people who explicitly consented to cross-posting were copied to RocketChat, and only when they contained the official conference hashtag. The cross-posting was handled using Zapier. The setup for this pipeline is saved in ACL's Zapier account and can easily be adapted for future conferences.

We also created a #live-microblogging channel for the micro-bloggers to discuss their experience and provide feedback. We’ve collected (anonymized) insights from this discussion and can share them with future publicity chairs.

Suggestions for improvement

Twitter

When posting announcements and award winners, we missed things because we were busy during the livestream parts of the plenary sessions. We strongly recommend that posts with important announcements, award winners etc. be prepared in advance and (to the extent possible) scheduled via tweetdeck, to reduce the load for publicity teams during the conference itself. For embargoed information, we encourage the PCs (or whomever knows it ahead of time) to write a textual version to share with the publicity chairs as soon as the information is announced for the publicity chairs to use to produce the tweets.

In the call for microblogging volunteers, it was not clear enough that this role did *not* come associated with reduced registration or other benefits apart from the reach we provided by boosting people's posts. We recommend either 1) associating a small benefit with microblogging (maybe it's possible to reduce their fee by a small amount?), or 2) making it explicitly clear from the beginning that this is *not* the same thing as being a conference volunteer. Some volunteers also didn’t understand that microblogging required registering for the conference. This should also be made clear in the call.

Weibo

During the conference, we invited three PhD students in Zhiyuan’s group to be on duty every 8 hours by turn. In this way, we can keep all important information like announcements, award winners to be posted in good time.

The Weibo official account and important posts should be also promoted via other social media platforms, such as Wechat groups related to the NLP community.

It is also important to keep Weibo active and attract followers by posting regularly when there are no conferences. Volunteers are also required as Weibo posters between conferences.

Press

Many of the journalists we contacted never got back to us. It is unclear how many of them actually attended the conference, and to what extent they were interested in the topics covered.

The lay summaries were a useful collection of texts accompanying the published papers. We encouraged authors to publish them elsewhere as well, e.g. on personal blogs or their institution's website. However, we suggest the following steps to make the process run more smoothly.

1. The invitation should be sent somewhat earlier than we did, perhaps about a week after the camera-ready deadline.
2. The summaries should be collected via a form, rather than over email. The form should ask the authors to indicate which area (or other organizing property, as chosen by the publicity chairs) their paper belongs to.
3. Authors should be instructed to include their name and contact information inside the pdf.
4. Further specifications could also be given regarding format and contents (Are pictures okay? Where should the title go?)
5. The invitation should clearly signal to authors that the deadline is firm (many, many summaries arrived after we had already assembled the pdfs by area for journalists).
6. Publicity chairs should consider encouraging authors to write lay summaries in non-English languages, especially if we know there will be a strong media presence for specific languages.

RocketChat

It is unclear to what extent conference participants were watching the #social-media-posts channel. However, since the technical setup exists and is transferable, it is little effort to reuse the pipeline in the future.

The number of tasks we booked on Zapier was overestimated. We calculated just under 10k tasks to be necessary and booked the next-highest plan with 20k tasks; because microbloggers posted shorter threads than anticipated, the number of tasks used on Zapier remained under 10k throughout the conference.

Unfortunately, out of the platforms used by our live microbloggers, only Twitter had an API that was compatible with Zapier and RocketChat at the time of the conference. Zapier seems to be working on including a Weibo API as well, which would be useful for future uses of this setup. A Mastodon API was set up, but none of the official microbloggers ended up posting on Mastodon.

Other suggestions

The virtual format of the conference came with a new set of challenges, both to the organizers and to the participants. Many participants shared opinions about the format on social media, and a debate has been going on about how to make the most out of virtual conferences.

1. We suggest installing a Virtual Social Chair for upcoming conferences. Their responsibilities should include:
2. Preparing blog posts for attendees with tips on how to interact with others at the conference
3. Planning a social program that can work in the virtual setting
4. Helping with designing the platform on which the conference is being held (for ACL 2020: RocketChat) in a way that is inclusive, welcoming, and not overwhelming
5. Planning and implementing a "buddy system" or mentoring program/big sibling program to make sure that first-timers and others who are unsure about the virtual format can have a way to interact that feels safe and that helps them find their way in the virtual conference format
6. Other efforts that facilitate networking, socializing, spontaneous and serendipitous encounters, and make the conference feel more interactive

The Virtual Social Chair should work closely with the infrastructure team, the diversity and inclusion team, and the publicity team. However, their tasks do not fall clearly into one of these categories, so it makes sense to view this as a separate role.

Esther Seyffarth is happy to volunteer for this role for either EMNLP 2020 or COLING 2020, and can be reached at esther.seyffarth@gmail.com.