With a delay of over two years, attendees of the bi-annual Language Resources and Evaluation Conference 2022 (LREC) finally arrived in sunny Marseille. After originally scheduled for May 2020, but then being canceled due to travel restrictions and nation-wide lockdowns, the organizers decided to revive the original conference location and agreed on the currently predominant setting of a hybrid conference with both physical and virtual attendance. As conference venue, the pompous Palais du Pharo was chosen, overlooking the charming ‘vieux port’ of Marseille, which made for an impressive backdrop during the numerous coffee breaks. Its central location also made it easy to move post-conference discussions to nearby restaurants and bars in the close-by port area.
Conference chair Nicoletta Calzolari worked tirelessly together with local organizers Frédéric Béchet and Philippe Blache to plan a worthy replacement for the cancelled 2020 edition, and even managed to bring the French ministry of Culture on board as a conference sponsor.
In total, over 1200 researchers decided to follow their invitation to a week at the French Riviera and the weather surely did not disappoint, with an average of 28°C and plenty of sunshine. As for the remote option, another 500 attendees managed to join in (and present at) the conference, which was organized professionally by Conference Catalysts, providing support in the form of a dedicated virtual platform and video upload management. Further live streams of the in-person sessions were provided by the conference organizers. In general, though, in-person and remote attendees had very little opportunity to interact between the two different groups at the event, which differs from recent hybrid experiences such as ECIR (read more about the setup at ECIR here). Although it should be said that the virtual platform was kept around even after the official event ended, with more opportunities for attendees to catch up on missed presentations or remote-only content.
During his opening remarks, ELRA (ELRA Language Resources Association) president António Branco gave a passionate speech about the importance of LREC establishing itself as a high-impact NLP venue, despite its unorthodox approach to acceptance rates. Historically, it boasts one of the highest (if not the highest) acceptance rate of any major conference in the field, with around 60-70%. Despite this – or potentially even because of it – LREC has shown tremendous growth over the years and is now a venue for resources and evaluation papers for a much greater variety of languages than other conferences. Ultimately, for the 2022 edition of LREC, just over 800 papers have been accepted and were presented at the conference.
Main Conference
Now in its 13th edition, LREC decided to embrace their focus area with a special call for “Multilingualism and Language Technology for All”, which was a prevalent theme throughout the conference. Not only did several of the workshops focus on lower-resourced languages, but several tracks during the main conference also addressed the surging number of submissions introducing further non-English resources. Historically, LREC does not have an overly large intersection with the retrieval community, but due to the recent popularity surge of NLP-based tools in the Search Stack, attendees could still freshen up on some of the latest trends in the area. Throughout the main conference, the programme consisted of four parallel presentation tracks, as well as the option to attend the poster area. Unfortunately, poster slots were also limited to the duration of a single session length (around 1.5 hours), which meant that one had to actively choose between attending presentations, or missing out on the opportunity to discuss a particular poster. However, the programme generally tried to span very opposite topics for any specific time slot which meant that most attendees would have a clear preference towards one of the sessions.
For all presentations at the venue, an author had to be present to ensure a “physical presentation”, which was supported by a local AV technician and two session chairs. The latter also included questions from the remote audience in the Q&A sessions after each presentation, but generally (remote) participation was remarkably low in the sessions that I attended.
Of particular interest might also be the recent changes to the ELRA, formerly known as the European Language Resources Association, now rebranding to the recursively titled “ELRA Language Resources Association”. During the ELRA general meeting, it was also discussed which implications this might bring for the future of its flagship conference, LREC. In particular, we might see future editions hosted elsewhere, with attendees favoring both a North American or Asian conference venue for the next edition. As of now, it is not decided where the 2024 conference will take place, and host bids are still warmly welcome. Finally, several participants also brought up the point that LREC must live its language diversity not only in the programme, but also in the support structure enabling authors from lower-resourced institutions to attend. Compared to structured support programs, such as the ACM SIG travel grants, ELRA currently does not have a fixed budget in place, which could be addressed in the upcoming years.
For the evening programme, the local organizers made sure that physical participants got to enjoy a part of local French cuisine: Not only was there plenty of French wine available to sample, but there was also plenty of authentic Marseille bouillabaisse available to try. For both the welcome reception, as well as the gala dinner, the terrace behind the conference venue stayed busy long after the official conference program had ended.
Keynote Talks
LREC also managed to impress with a strong speaker lineup for its keynote and invited talks.
Starting the conference off on Tuesday with the first keynote was Julia Parish-Morris, who took her listeners on a journey through the world of NLP in clinical settings, particularly detailing the importance and diversity in creating language resources for (and with) people suffering from autism. Especially from a neurological perspective, their inclusion in the representation matters, and helps to pave the way to true equality.
On Wednesday, Emmanuel Dupoux gave a great overview of recent advancements in his keynote on what he calls “Textless NLP”. Language processing from audio data has become increasingly more important (a fact that could also be observed by surveying other works at the conference), and direct processing without intermediate automatic speech recognition (ASR) steps would greatly improve our ability to pre-train large language models directly on raw audio data.
Philippe Boula de Mareüil brought with him a series of linguistic maps of France for his invited talk on the linguistic diversity within a single language (French). He also reminded attendees to always pay attention not only the linguistic differences, but also differences in the speaker population, especially for minority speaker groups. Also focusing on the differences between different speaker profiles was José Deulofeu, who particularly investigated the origin of spontaneous differences in oral versus written dialogue forms.
For many, however, a clear highlight of the conference was the Antonio Zampolli award winner, Steven Bird, for his work on endangered and low-resource languages. In light of his award, he took the opportunity to give a full 30 minute acoustic guitar concert in the auditorium, probably a first for any award recipient. (You can listen to a video recording of the concert here). His award speech consisted of a collection of five different stories from his life. He concluded with encouraging word on embracing linguistic diversity as an opportunity, instead of seeing it as a “problem to solve.”
Workshop and Tutorial Days
Of course, with an additional two years of pent-up submissions, the conference program was packaged, boasting a grand total of 32 (!) workshops and 9 tutorias over the span of three days, padding the three days of main conference tracks. For many, the registration process for workshops and tutorials was experienced as a bit unfortunate, as separate (costly) sign-up fees were charged for each individual workshop or tutorial, even though the registrations were not even checked in the end. Further confusion was added by the initially incorrectly charged prices for students, which made the entire process unnecessarily tedious.
As always, though, workshops were still a great place to interact with a more focused group of researchers with similar interests. I found the Perspectives workshop to be quite interesting, which dealt with the incorporation of subjective user perspective into the collection and annotation process. This is strictly speaking not limited to the NLP community, and will hopefully see a wider surge in popularity in the near future, especially with an increasing focus on (manually) curated datasets as a gold standard way of evaluating methodological results in papers.
Conclusion
Overall, though, LREC has been a fantastic experience, being one of the first larger conferences to open its physical doors after an unexpected covid break. As organizer Nicoletta Calzolari put it, organizing a hybrid conference almost feels like planning two separate conferences at the same time. This will become problematic, unless we dedicate more resources (and people) to the planning of future events, especially if we expect hybrid setups to stick around for the foreseeable future.
A less comfortable, but nonetheless important issue, will also be the question on how we deal with seasonal infections (Covid or other) in the future. During and after the conference, several instances of infections became known, and no recommended or enforced mask requirements at the venue meant that infections could freely spread throughout the week. How we can balance the amount of required medical rules with the urge to maximize face-to-face interactions during physical events will have to be addressed in future planning committees, else we face a continued risk of re-infections at each future event.
LREC especially can be commended on their effort of making presentations and conference material available online, free of charge. In fact, you can already take a look at any of the conference papers on the website.