We seek a postdoctoral researcher (starting date: January 1st, 2027) to conduct the first systematic investigation of reasoning circuit structure in compact multilingual language models. This is a bilateral DFKI-Inria collaboration project funded by BMFTR (SLIMS), bridging mechanistic interpretability and multilingual NLP, addressing a critical gap: existing circuit analysis focuses on large English-centric models, leaving the question of cross-lingual and cross-scale circuit transfer largely unexplored.
You will work closely with the ALMAnaCH team at Inria Paris (Prof. Benoit Sagot, Dr. Djamé Seddah) on controlled validation using backdoored model suites as ground truth. The project includes planned bilateral research visits and joint workshops, enabling direct collaboration with leading researchers in multilingual modeling and interpretability methodology.
You will apply the full mechanistic toolkit (activation patching, sparse autoencoders, natural language autoencoders, transcoders) to understand whether reasoning circuits survive compression and translate across typologically diverse languages. Your work will directly inform the design of more capable multilingual models and advance interpretability methods for linguistically diverse systems. We value rigorous methodology and open discussion of results, regardless of whether they confirm initial hypotheses or reveal surprising findings.
Importantly, you will work as part of a fully integrated team composed of members of another BMFTR-funded project, SAgA, ensuring that your circuit-level discoveries directly feed into safe agentic system design, creating a tight feedback loop between mechanistic analysis and safety architecture.
The position is embedded in the Multilinguality and Language Technology (MLT) group under the direction of Dr. Marius Mosbach, with broader departmental support from Prof. Kristian Kersting (SAINT deparment, Darmstadt) and Prof. Verena Wolf (NMM department, Saarbrücken); and in close colaboration with the ALMAnaCH team at Inria Paris (Prof. Benoit Sagot, Dr. Djamé Seddah); the primary supervisor of the position is Dr. Simon Ostermann; the position will be situated in his team "Efficient and Explainable NLP"
Apply circuit analysis methods to identify and characterize reasoning circuits in small multilingual models
Required Background
Desirable Qualifications
The German Research Center for Artificial Intelligence (DFKI) has operated as a non-profit, Public-Private-Partnership (PPP) since 1988. DFKI combines scientific excellence and commercially-oriented value creation with social awareness and is recognized as a major "Center of Excellence" by the international scientific community. In the field of artificial intelligence, DFKI has focused on the goal of human-centric AI for more than 35 years. Research is committed to essential, future-oriented areas of application and socially relevant topics.
DFKI encourages applications from people with disability; DFKI intends to increase the proportion of female employees in the field of science and encourages women to apply for this position.
© DFKI,Â