ZurichNLP #19

Sina Ahmadi (UZH): Language beyond the Standard: NLP for Low-Resource Varieties

NLP has made remarkable progress, yet these advancements remain unevenly distributed, with low-resourced languages still significantly underserved. This talk explores how we can broaden the scope of language technologies by developing computational models that effectively support languages with limited data, documentation, or digital presence. I discuss core challenges, including data scarcity, orthographic and linguistic variation as well as emerging strategies for data creation, model adaptation, and community-driven resource development. By addressing these issues, we can move toward more inclusive and equitable language technologies.

Barna Pasztor (ETH): Sample-efficient dataset collection for Reinforcement Learning from Human Feedback with active learning

Increasing accessibility of RLHF via reducing annotation costs and identifying valuable datapoints to training reward models and DPO fine-tuning.