Exploring the Frontiers of Language and Speech: Highlights from the 2025 ELLIS NLP & Speech Recognition Workshop

This year's ELLIS Workshop on Natural Language Processing and Speech Recognition took place on June 4, 2025, in a fully online format. The event was jointly organised by Josef Sivic from the ELLIS Unit Prague and Rostyslav Hryniv from ELLIS Associate Unit Lviv, and featured a rich program of invited talks and interactive discussions.

The workshop attracted strong interest, with 160 overall registrations, 40-60 joining live on Zoom, and already around 200 YouTube stream views. Additionally, a local session at the Ukrainian Catholic University gathered around 20 undergraduate and PhD students who also joined the online workshop via a joint Zoom session.

The program consisted of six invited talks by experts in the field:

Iryna Gurevych, one of the directors of the ELLIS NLP Program, opened the workshop with a talk on spotting and debunking misleading content - a timely and critical topic in today’s information landscape.

ELLIS Fellow Barbara Plank explored reasoning, ambiguity, and control in Large Language Models (LLMs), and presented probing techniques for understanding what these models actually "know".

Mykola Trokhymovych (PhD Student at University Pompeu Fabra) addressed how NLP can support accessibility, fairness, and knowledge integrity in collaborative platforms such as Wikipedia.

ELLIS Member Alessandro Raganato discussed methods for blending different concepts specified by natural language in the context of image generation.

Oldřich Plchot (Researcher at Brno University of Technology) presented work on combining large language and speech models to enable smooth and consistent dialogue.

Yehor Smoliakov (project https://ua-lawyer.com) gave an overview of the current state of Ukrainian Speech-to-Text and Text-to-Speech technologies, including datasets created by the research community.

The workshop also featured lively Q&A sessions, with questions ranging from how to define and measure misinformation, to how LLMs deal with ambiguity, and how to transfer research results from English to low-resource languages in both NLP and speech. Participants also asked for advice on applying to the ELLIS PhD Program, making the workshop not only a venue for sharing research but also for supporting the next generation of scholars.

Overall, the event showcased current advances in NLP and speech recognition, and reflected the growing global collaboration within the ELLIS community.