SwissText 2022 was organised jointly by SUPSI, SwissNLP and  the  ZHAW Centre for Artificial Intelligence (CAI). The annual conference is a forum for researchers and practitioners in NLP to meet and discuss. After two online editions, it was finally held again as a physical conference at the SUPSI East Campus in Lugano.

The pre-conference day included interactive workshops on keyword extraction from scientific documents, Swiss German (Speech-to-Text and lexical normalisation), and NLP for Insurance, plus a co-located event on a programme for upskilling linguists for technical professions.

During the two main conference days, there were a total of 18 presentations organised into thematic tracks such as “Speech-to-Text and Swiss German”, “Legal Applications”, “Generation and Parsing” and “Business Applications”. The newly created Junior Track, designed to provide a platform to young researchers, featured six insightful presentations.

Two interactive events made sure that participants were able to further showcase their work and provided valuable networking opportunities:
first, the exhibition on Thursday afternoon featured 21 research posters, 4 system demonstrations, as well as 11 booths by the conference sponsors and affiliated academic institutions.

Second, in the “Battle of NLP Ideas” on Friday, participants discussed in small groups to come up with ideas for new NLP projects. In subsequent rounds, groups were merged in a pyramid fashion and selected the most promising suggestions before presenting them in a plenary meeting where the audience could award votes. The three winning ideas were on synthetic data generation, identifying bot-generated content and data anonymization for NLP. All participants were then able to sign up for those ideas where they are interested in a follow-up meeting.

The rich and varied programme was complemented by three keynotes by renowned experts:

Raul Rodriguez-Esteban, Senior Principal Scientist at Roche, discussed Quantitative Social Media Listening, a relatively new trend in health care where social media are used to identify not yet documented symptoms of a disease or populations with unmet medical needs, or find the best location for a clinical trial.

Google’s Enrique Alfonseca introduced their ongoing work on integrating structured knowledge into large language models. This is important in the context of enhancing the reasoning capabilities of these models and creating more factually correct responses.

Marco Passarotti, professor at the Catholic University of Milan, talked about the benefits of Linked Data Interoperability when creating language resources, which he illustrated with his ongoing ERC project «LiLa: Linking Latin».

Overall, it was a very successful conference and the beautiful Ticino weather was a big bonus! The slides and recordings of the talks will be made available on the conference website in the coming weeks. SwissText 2023 will continue the journey across Switzerland’s language regions: it will be held at Haute École Arc Ingénierie (HE-ARC) in Neuchâtel.