22nd Belgium NLP Meetup

With the freezing temperatures outside, it's not just our NLP models that could use some extra layers. Fortunately on Tuesday, February 27th, you can warm up in the company of fellow NLP enthusiasts and join us for our 22nd Belgium NLP meetup. We're meeting at the BeCode office, conveniently located above the Central train station in Brussels. Two presentations are already on the agenda: François Remy and Pieter Delobelle will delve into their latest RobBERT models, while Luna De Bruyne will shed light on her innovative work in multilingual emotion detection. Stay tuned for the announcement of our third and final speaker.

As always, doors open at 7 pm, with talks kicking off around 7:30 pm. We look forward to seeing you in Brussels!

Tik-to-Tok: Turn your monolingual LLMs into multilingual mavens
François Remy (UGent) & Pieter Delobelle (KU Leuven)
Unlock the potential of AI in any language with ease! In this talk, you will learn the secrets of efficient language model conversion, the process of adapting small and large language models to new languages. Pieter and François will explain the technology behind the RobBERT-2023 models, and some newer advances they have made since then.

Emotions without borders: Challenges in multilingual emotion detection
Luna De Bruyne (University of Antwerp)
Emotions have attracted a lot of attention in psychology and linguistics, and since the past decade also in the field of NLP. Research on emotion detection has mainly focused on English, but with the emergence of large language models, the interest in multilingual approaches to emotion detection has increased. Meanwhile, state-of-the-art research in psychology has developed new theories about emotion, claiming that emotions are not universal: neither in conceptualization, nor in expression. This might have consequences for how emotion detection models work. In this talk, Luna will describe the challenges in multilingual emotion detection and discuss some ongoing and future work concerning this topic.

Honey I Shrank the Llama
Maarten Mortier (Tekst.com & Henchman)
An overview of techniques to host data- and space-hungry LLM's in real-life settings. How do you perform fine tuning for thousands of customers? Where does quantisation and pruning happen? How many acronyms can you *really* fit in a presentation without losing the audience? Find out with Maarten, who is one of the innovative brains at Henchman and Tekst.com.