Denna sida på svenska

Research group Natural Language Processing Research Group

The Natural Language Processing Research Group develops, applies and evaluates NLP methods, in particular involving large language models, across various domains. We focus on topics such as privacy, explainability, and domain adaptation.

HPV-16 cells - a high-risk type for cancer.

The research group works with different types of data, healthcare data is one example. Photo: National Cancer Institute/Unsplash.

The Natural Language Processing Research Group carries out research on topics concerning methods for processing, modeling and analyzing text, including large language models (LLMs). We are motivated by real-world applications of Natural Language Processing (NLP) in domains such as healthcare, education, and security.

We have built up extensive expertise in clinical NLP for analyzing healthcare data and, to that end, have a research infrastructure called Health Bank. Clinical NLP methods enable automatic large-scale analysis of healthcare data and are valuable for improving healthcare, for example by building clinical prediction models that incorporate information from clinical notes. We explore how LLMs can be used in healthcare and apply domain adaptation for creating clinical language models, especially using privacy-preserving NLP methods (such as de-identification or synthetic training data).

Another focus application area is education, where we are interested in using pre-trained language models for different educational use cases – automated essay scoring, question answering/generation and educational content recommendation – to enhance teaching and learning processes. To enable development of intelligent and adaptive learning systems, we explore techniques such as retrieval-augmented generation (RAG) and tool-augmented generation (TAG).

Security is another application area where our focus is on detection and analysis of online harms such as hate speech, threats, and violent extremist content. We are also interested in threat assessment of written communication and to determine the seriousness of threats. We host the European Online Hate Lab, a hub for researchers and organisations that detect and analyse online hate.

Furthermore, explainability is critical for the development of trustworthy AI, therefore we develop methods for explainable NLP, especially in relation to LLMs. We focus especially, but not exclusively, on developing NLP methods for the Swedish language.