SwissText 2023 overview

SwissText 2023

---

The 8th edition of SwissText took place from June 12 to 14, 2023 in Neuchâtel.

It was organised by SwissNLP jointly with HE-ARC and  the  ZHAW Centre for Artificial Intelligence (CAI).

Programme Overview

The pre-conference day included interactive workshops on Text Mining and Biodiversity, Keyword Consolidation and a tutorial on the OpenDataflow engine as well as two shared tasks, Detecting Greenwashing Signals and the Swissdox Hackathon.

During the two main conference days, there were a total of 16 presentations organised into thematic tracks such as “LLM Applications” or “Summarisation and Translation”. The Junior Track was held for the second time and showcased the work of five young research teams.The interactive exhibition on Tuesday afternoon featured 18 research posters and booths by sponsors as well as affiliated academic institutions.

Social Event: Culinary Cruise on Lake Neuchâtel

A special highlight was the Social Event on Tuesday evening, which took place on a boat and gave participants the opportunity to enjoy scenic views of Lake Neuchâtel while tasting an apéro of Gâteau de Vully and local wines.

Third Battle of NLP Ideas: Validation of Asylum Decisions

The “Battle of NLP Ideas” was held for the third time. After three rounds, the proposal by Alexandros Paramythis to validate (and help challenge) asylum decisions using NLP was declared the winner in an audience vote – congratulations! Details of all six finalist ideas can be viewed here: https://drive.google.com/file/d/1_GcMIZXDc8cTxsFkJxAt110SzbO686gw/view

Third SwissNLP Award goes to STT4SG-350

The third SwissNLP Award was presented to the creators of the STT4SG-350 corpus, the largest Speech Translation Corpus for Swiss German to date. The three universities behind the project, FHNW (Fachhochschule Nordwestschweiz), ZHAW (Zurich University of Applied Sciences) and UZH (University of Zurich) created a balanced corpus of nearly 350 hours of Swiss German audio recordings and the corresponding Standard German texts. STT4SG-350 is available for research and commercial purposes.

Five Exciting Keynotes

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

• Anna Rogers, Assistant Professor at the IT University of Copenhagen, raised a number of important concerns about Large Language Models in her talk Towards Better Data Governance for Large Language Models. She discussed the current data and privacy policies of commercial LLMs and presented the BLOOM large science project, which takes an open and transparent approach to data governance.

• Jacques Savoy from the University of Neuchâtel talked about text classification with style, providing a comprehensive overview of the different tasks (such as gender and age profiling and author verification) and methods.  

• Diego Antonini, Research Scientist at IBM, discussed Efficient Machine Learning in Low-Resource and Highly-Specific Domains, where the challenge is to create weight-efficient models with much smaller sizes and shorter inference times that can be used in low-data regimes and can be deployed on-device.

• Claudiu Musat, Research Manager at Google, introduced Digital Ink – Modern Processing of the Oldest Textual Form, i.e. the task of recognising and synthesising handwriting on digital surfaces. He highlighted some of the challenges of this field as well as similarities between processing Digital Ink and non-standardised languages such as Swiss German.

• Silvia Quarteroni is the Chief Transformation Officer of the Swiss Data Science Centre. She presented her comprehensive experience of Bringing Natural Language Processing Applications to Swiss Organisations, giving an overview of the most common use cases and recounting some success stories. It became clear that not all NLP technologies are alike when it comes to the challenge of integrating them into an existing IT ecosystem.

Thanks to the local organisers at HE-ARC!

The conference was a great success due to the monumental effort of the local organising team! The slides and recordings of the talks will be made available on the conference website in the coming weeks.