2021Q3 Reports: SIGTYP
Summary
SIGTYP is ACL's special interest group on linguistic typology.
President: Ekaterina Vylomova
Secretary: Ryan Cotterell
At Large: Eitan Grossman, Edoardo M. Ponti, Silvia Luraghi, Alexis Palmer
Membership
The SIG was officially approved on Dec, 28 2019. In July 2021, the total number of members has reached 380. We are planning to hold a membership drive to further promote growth within the SIG.
Workshop
In summer 2019 we organized the first workshop on typology for polyglot NLP (co-located with ACL 2019). In total, 48 attendees registered for the workshop (excluding organizers and keynote speakers). In autumn (November) 2020, we ran the second (virtual) workshop on computational approaches to linguistic typology (co-located with EMNLP 2020). In total, ~50 attendees registered for the workshop. In summer (July) 2021, we organized the third (virtual) workshop on computational approaches to linguistic typology (co-located with NAACL 2021). We developed our own virtual infrastructure (https://sigtyp.github.io/ws2021-schedule.html), ran several sessions during 24 hours to allow members from different time zones attend any session they prefer. According to our records, in total ~130 unique participants attended the sessions.
Organizers of 2021 workshop:
Ekaterina Vylomova, Elizabeth Salesky, Sabrina Mielke, Gabriella Lapesa, Ritesh Kumar, Harald Hammarström, Ivan Vulić, Anna Korhonen, Roi Reichart, Edoardo Maria Ponti, Ryan Cotterell
Keynote Speakers:
Claire Bowern, Miryam de Lhoneux, Johannes Bjerva, David Yarowsky
In 2021, SIGTYP offered a shared task on the robust prediction of language ID from speech. In the task, we addressed one major issue: for many low-resource and endangered languages, only single-speaker recordings may be available. Therefore, such conditions require domain and speaker-invariant language ID systems. We asked the participants to build systems that will be trained on largely single-speaker speech from one domain, but evaluated on data in other domains recorded from speakers under different recording circumstances, mimicking more realistic low-resource scenarios. In total, 3 teams participated in the task. Th results demonstrate that the task is challenging.
Other Activities (Online)
SIGTYP website and logo
We developed SIG’s website ([1]). It is constantly being updated with new information on workshops, shared tasks, members, and other information. We also designed a group’s logo: [2]
SIGTYP Lecture Series ([3]) Every week we invite a speaker either from NLP or linguistic typology to present their research. We pre-record the talk in four ~15min parts and then play them having live discussions after each. For this purpose, we created our own Youtube and Bilibili (China) channels: Youtube: https://www.youtube.com/channel/UCaSWMbnmduXYlbWGEWLedww/about Bilibili: https://space.bilibili.com/1055445444
SIGTYP digest ([4])
Each month we invite members of the community to submit short abstracts or summaries of their recent papers to our monthly newsletter. This allows keeping track of the progress in the field and promoting everyone’s work.
Twitter account ([5])
We engage more members by keeping our Twitter account constantly updated with retweets of recent papers, talks, and other materials on linguistic typology, multilinguality, and low-resource NLP.
Mailing Lists
We created organizational structure for SIGTYP, e.g. Google groups for: 1) SIGTYP members; 2) SIGTYP Exec; 3) SIGTYP shared task organizers.
Elections
As secretary, Ryan Cotterell will be organizing the elections over the coming months. He is going to follow Garrett Nicolai's procedure at SIGMORPHON for remote voting.