NLP research group @IDSIA

Glad to report that our paper received a "SAC Highlights Award"!

Tokenization and Representation Biases in Multilingual Models on Dialectal NLP Tasks Dialectal data are characterized by linguistic variation that appears small to humans but has a significant impact on the performance of models. This dialect gap has been related to various factors (e...

Presenting at #EMNLP2025 in a moment, session on "Multilinguality and Language Diversity 2" (A301). Our paper on Tokenization Fairness: arxiv.org/abs/2509.20045

The project is funded by a large SNF grant, awarded to Dr. Fabio Rinaldi (IDSIA USI-SUPSI), Prof. Michael Krauthammer (UZH), Prof. Jean-Louis Raisaro (CHUV). Participating hospitals are EOC (Ticino), CHUV (Lausanne), and USZ (Zurich).

The project leverages AI technologies to enable a safe sharing of de-identified medical documents, to accelerate medical research. Additionally, as a proof-of-concept, it will use the de-identified documents to investigate adverse drug reactions in Oncology and discharge summary generation.

M2P2 stands for Medical, Multilingual and Privacy-Preserving Natural Language Processing in the clinical domain.

Yesterday we had a great kick-off meeting of our M2P2 project, a new Swiss-wide initiative for clinical text processing recently approved by the Swiss National Science Foundation!
nlp.idsia.ch/PROJECTS/M2P...

This is the inaugural post for the idsianlp account on Bluesky.