The rapid advancement of generative language models (LMs) has significantly influenced the creation of high-quality texts, aiding both expert and non-expert users in various tasks. However, the potential misuse of these LMs for generating fake news, product reviews, and extremist content has raised concerns. Machine-generated text (MGT) detection has emerged to develop benchmarks and detectors to counter such misuse and become pretty popular due to the increasing difficulty of identifying the generated content.
This thesis focuses on MGT detection for Norwegian (and other Scandinavian languages, which can be discussed with the supervisors). During this masters project, you will gain and improve your experience with (i) automatically collecting and curating data, (ii) utilizing generative LMs in different setups, and (iii) training and evaluating MGT detectors.
Prerequisites: Good knowledge of Norwegian and Python.