Wykład mistrzowski dr. Michała Ptaszyńskiego
Serdecznie zapraszamy Państwa na wykład mistrzowski dr. Michała Ptaszyńskiego (Kitami Institute of Technology) o tytule: Beyond English: How Language Similarity Predicts the Performance of Multilingual AI.
Wykład odbędzie się 7 maja 2026 roku o 10.00 zdalnie za pośrednictwem aplikacji MS Teams.
Abstrakt wykładu
When building AI tools for new or low-resource languages, developers usually default to training their models on English. However, this intuitive approach often leads to sub-optimal results. In this lecture, I will explore why linguistic similarity is the true key to unlocking successful cross-lingual AI.
I will introduce qWALS, a novel metric that calculates the grammatical, phonological, and semantic proximity between over 2,500 languages. By evaluating multilingual transformers across diverse NLP tasks, I will demonstrate how choosing a similar source language drastically improves zero-shot model performance compared to simply using English.
I will propose the idea that in multilingual tasks language models behave as polyglots, by exploiting knowledge from other languages.
The proposed qWALS language similarity score is a data-driven framework for building more effective and equitable multilingual technologies.
Wykład będzie prowadzony w języku polskim.
Biogram
Michał Ptaszynski received the master’s degree from the University of Adam Mickiewicz, Poznan, Poland, in 2006, and the Ph.D. degree in information science and technology from Hokkaido University, Japan, in 2010. From 2010 to 2012, he was a JSPS Postdoctoral Research Fellow with the High-Tech Research Center, Hokkai-Gakuen University, Japan. Later, he joined Kitami Institute of Technology, as Assistant Professor, then Associate Professor. Currently, he is a Full Professor with Kitami Institute of Technology. He is the founder of the fields of Automatic Cyberbullying Detection and Ainu Language Processing. His research interests include artificial intelligence, and natural language processing, especially of low-resource, indigenous, and endangered langauges. He is a member of AAAI and ACL, and a Senior Member of IEEE.
