• lm
• cross-lingual transfer
• differential privacy

This thesis evaluates biomedical Language Models (LM) in multi-lingual and privacy-preserving scenarios. This work has two goals: first, to assess multi-lingual GPT models using various cross-lingual transfer experiments, and second, to evaluate the implications of Differential Privacy on synthetic datasets generated by a fine-tuned LM using biomedical data. We particularly aim to explore the dual effect of Differential Privacy in this context. On one hand, we are interested in its contribution towards enhancing privacy, while on the other hand, we aim to understand its impact on the practical utility of synthetic datasets derived from this LM. This thesis was developed as part of the DataTools4Heart European project during an internship at Translated.