By Trine Jensen, AAU Communication & Public Affairs

The Data Science Investigator grant is intended for excellent and independent Associate Professors (or equivalent) who have the ambition and potential to rise to the highest international level. The purpose of the grant is to support ambitious projects rooted in data science and computational science, thereby strengthening the quality and size of the Danish academic environment for data science research and education. The aim is to produce more skilled candidates in data science to meet the high demand for such competencies from all sectors of society.

The grant is a significant recognition that allows Johannes Bjerva to focus on establishing an entirely new field of research.

"Receiving the Novo Nordisk Foundation's Data Science Investigator grant is a fantastic recognition of my research in Natural Language Processing (NLP) and a significant step in my career as a research leader. It gives me the opportunity to focus on establishing a new field of research and strengthening Denmark as an international hub for security in generative AI and large language models (LLM)," says Johannes.

About the project: LM2-SEC: Linguistically Motivated Language Model Security

Johannes' research project builds on Natural Language Processing (NLP) and integrates knowledge from Computational Linguistics (CL) as well as from early research in NLP security. The research is centered around a new data resource that allows for the investigation of signals sent to, from, and even between large language models (LLMs). LM2-SEC aims to distinguish between malicious and benevolent influences on these signals.

“Our hypothesis is that linguistic regularities in these signals can be isolated and rectified. The project lays the foundation for a leading research environment in NLP security, leaning on linguistics, and contributing to filling the gap of humanities-oriented Data Science in Denmark,” explains Johannes.

Johannes' most significant scientific contribution has been incorporating linguistic information into NLP, thereby improving the state of especially resource-poor languages. Specifically, his research group's contributions over the past three years have highlighted the relevance of linguistics in NLP, showing that linguistically informed approaches to language sampling can have significant effects on the performance of NLP models.

The significance of the grant for NLP research

This is the first time that the Novo Nordisk Foundation has awarded a major grant to this field, and it provides an opportunity to research the latest challenges in NLP.

"The grant allows me and my team to tackle some of the latest challenges in AI security and NLP. We will systematically explore how knowledge from linguistics can be used to detect security threats against large language models (LLMs). My hope is to create a new, scalable, and theoretically grounded approach to NLP security. This is particularly relevant if one wants to implement language models within the healthcare sector. At the same time, the grant will establish a leading research environment in Denmark with strong international collaborations, including with NVIDIA," says Johannes.