230022 Linguistic perspectives on Large Language Models (S) (WiSe 2024/2025)

Large Language Models (LLMs) like GPT-3 or GPT-4 have greatly advanced the state-of-the-art in computational linguistics and natural language processing. For the first time in the history of this field, there seems to be a technology that comes close to human language use in various kinds of situations and tasks. For this reason, LLMs have recently received interest from linguistic research: linguistic researchers are currently debating to what extent LLMs can be seen as a model of the human ability to process language and use it for communication. In this research-oriented seminar, we will first cover some computational basics of LLMs and then dive into this debate: what do LLMs know about human language and how can we test this? How can we use insights and experimental designs from linguistic research to study LLMs? And maybe: how can we use LLMs to learn something about human language?

Requirements for participation, required level

This class targets an interdisciplinary audience of computationally-minded and linguistically-minded students, including PhD students.

Basic knowledge of (python) programming and machine learning will help but is not strictly necessary.