Postdoctoral Position in NLP Post-Training for Cultural Alignment and Preference Optimization

The AarhusNLP Group at the Center for Humanities Computing, Aarhus University, invites applications for multiple three-year Postdoctoral Researcher positions. Starting date: June 1 2025 (or as soon thereafter). The successful candidate will be physically based with the AarhusNLP Group on projects funded by the Danish Ministry of Digital Affairs’ AI Initiative. This initiative stems from the 2024 national AI Strategy and is carried out in collaboration with major Danish universities, including the University of Copenhagen, the University of Southern Denmark, and the Alexandra Institute. 

Place of work:  

The successful candidates will work primarily with the AarhusNLP Group at the Center for Humanities Computing, Aarhus University. The Center offers flexible working arrangements, including options for remote work and adaptable working hours, to support work-life balance and accommodate individual needs. Candidates will be expected to participate regularly in on-site activities, fostering a collaborative and engaging research environment. 

The postdocs will work on cultural alignment and preference optimization of large language models (LLMs) for mid to low-resourced languages. The overarching goal is to develop advanced post-training techniques that align pre-trained transformer-based LLMs with cultural values and linguistic conventions. By focusing on low-resource languages, this project addresses challenges such as data scarcity, cultural sensitivity, inclusivity, and the need for robust preference optimization methods that go beyond standard fine-tuning. 

Key research objectives include: 

• Developing Efficient Post-Training Techniques: Exploring and designing post-training strategies (e.g., Supervised Fine-Tuning, Reinforcement Learning, or novel alignment protocols) targeting Danish and other Nordic languages. 
• Cultural and Ethical Alignment: Embedding Nordic cultural norms, ethical considerations, and societal values into LLMs, ensuring respectful and context sensitive language generation. 
• Preference Optimization and Evaluation: Enhancing model performance on non-differentiable properties, including user satisfaction, cultural appropriateness, and domain-specific quality measures (e.g., information retrieval or question-answering for Nordic use cases). 

Qualifications:

• A Ph.D. or similar qualifications in Computational Linguistics, Computer Science, or a related field (e.g., cognitive science), specializing in Natural Language Processing, Machine Learning, or a similar area. 
• Expertise in large language model architectures and training paradigms (transformer models, fine-tuning strategies, RLHF, etc.). 
• Interest in or prior experience with multilingual or low-resource NLP 
• Programming skills in Python and proficiency in deep learning frameworks (e.g., PyTorch, TensorFlow) and their application in high-performance computing environments. 
• An established or emerging publication record in NLP, ML, or related fields. 
• Excellent written and oral communication skills in English. Knowledge of Danish or another Nordic language is beneficial, though not required. 

The position: 

As part of this project, you will contribute to advancing responsible AI by ensuring language models align with societal values, user needs, and ethical considerations in real-world applications. 

The successful candidates will be responsible for tasks including, but not limited to: 

• Developing and evaluating efficient post-training techniques for LLM alignment.
• Conducting research on embedding cultural norms and ethical values into language models. 
• Implementing preference optimization methods to enhance model performance on qualitative metrics. 
• Collaborating with project partners at other Danish universities. 
• Disseminating research outcomes through publications, presentations, and public engagement.

We strongly encourage applications from women and other underrepresented groups in AI and NLP.

For more information on the School of Culture and Society, please visit School of Culture and Society - Aarhus University

Advert information