Guest Lecture: Dr. Milena Belosevic (Bielefeld) — ‘(How) does context help language models identify linguistic patterns in German?’

Place: B 211

Abstract: Despite the almost human-like language capabilities of language models, many aspects of their linguistic behavior remain underexplored, such as their ability to engage with the kinds of context that humans routinely consider when identifying and interpreting linguistic units. This talk presents three case studies in which German language models are fine-tuned to explore whether and how varying the context type shapes their ability to identify linguistic patterns in naturally occurring human-human conversations, text corpora, and multimodal settings. Rather than promoting reliance on model performance, the results emphasize the importance of human interpretation and oversight in studying linguistic patterns.