I am a first-year PhD student at Saarland University and the German Research Center for AI (DFKI)One of the largest non-profit AI research institutions in the world, with its main site in Kaiserslautern and further sites across Germany, including Saarbrücken.. My research focuses on interpretability and efficiency in multilingual language models, supervised by Dr. Simon Ostermann and Prof. Dr. Josef van Genabith. For an overview of my work, see my Google Scholar profile or scroll further down this page. I am also contributing to the pre- and post-training of SOOFI, the first German-centric open-source foundation model, focusing on multilingual reasoning data generation.
Currently, I am a RIKEN-associated student trainee at Yanaka's lab at the University of Tokyo, where I am spending 6 months, until March of 2026, focusing on compositional representational steering and mitigating model forgetting. I am also organizing the CLaS shared task on language steering at SemEval 2027.
Along the way, I have also been fortunate to attend the ALPS NLP SchoolAdvanced Language Processing School, a week-long NLP school held in the French Alps, co-organized by Grenoble Alpes University and Naver Labs Europe. at Grenoble University and the MLSSMachine Learning Summer School, a long-running series of intensive graduate-level schools on statistical machine learning, running since 2002. summer school at Columbia University, as well as spend a month as a research visitor at KInITKempelen Institute of Intelligent Technologies, Slovakia's first independent private AI research institute, based in Bratislava. in Slovakia. I have also had the pleasure of co-organizing three events for the NLP community: TaCoSTagung der Computerlinguistik-Studierenden, Germany's annual student conference on computational linguistics., the 33rd Student Conference on Computational Linguistics, the LT-Bridge Winter SchoolA DFKI-led project on multilingual and efficient NLP technologies, spanning chatbots, machine translation, and low-resource language processing. on parameter-efficient methods for NLP, and the 1st Workshop on Advancing NLP for Low-Resource Languages (LowResNLP), co-located with RANLP 2025.
I completed my Master’s degree in Language Science and Technology at Saarland University, supported by the DAADDeutscher Akademischer Austauschdienst, Germany's national agency for funding international student and researcher exchange. and Erasmus+ scholarships. As part of my master's studies, I also spent a semester at the Center for Mind and Brain Sciences at the University of Trento (Italy). During this time, I worked as a Research Assistant in the Multilingual Language Technology Lab at DFKI, collaborating closely with my current supervisors.
Before moving to Germany, I earned my Bachelor's degree in Linguistics and Pedagogy at Shevchenko Transnistrian State University (Transnistria)A breakaway territory on the eastern bank of the Dniester, formally part of Moldova but functioning as a self-governing, internationally unrecognized state since 1990.. As part of my undergraduate studies, I completed an exchange semester at Murray State University (USA) through the fully funded Global UGRADGlobal Undergraduate Exchange Program, funded by the U.S. Department of State, bringing undergraduates from underrepresented countries to study for a semester in the US. program.
My early fascination with the intersection of mathematics, technology, and language naturally steered me toward Artificial Intelligence and later to Natural Language Processing. This journey has been both challenging and rewarding, and I am excited to continue exploring the inner workings of multilingual models and contributing to their development.
My research goal is to move from empirical steering methods toward a principled theory of controllable representations, using multilinguality as a lens for structure that stays invisible in English-centric analyses. I organize this around three pillars.
A rough story of what I've worked on so far during my PhD, in the order it happened.
A large-scale study of language-specific neurons across 21 languages and 4 base models, investigating the influence of activating and deactivating language neurons on language fidelity and downstream task performance (QA, MT, MC, and NLI). (!) Progressive neuron deactivation reveals internal language "fallback" mechanisms.
read the paper →CLaS-Bench, a benchmark for evaluating language forcing in LMs across 32 languages. Methods evaluated: DiffMean vectors, MLP neurons, PCA subspaces, LDA vectors, SAEs, and probes. (!) Simple DiffMean vectors tend to outperform all the other approaches and prompting. (?) Language identity as a linear direction in representation space?
read the paper →Script also seems to be on a linear axis and is gated by a small set of language-agnostic attention heads that causally control it, all concentrated in later layers. (!) Script-mediating attention heads generalize cross-lingually; identified on Hindi, they causally influence script choice in Russian, Japanese, etc.
read the paper →Position: simple linear edits at the activation level look like a solid alternative to "heavier" adaptation methods (fine-tuning, RL, or PEFT) for steering model behavior. (!) We provide a set of functional criteria for adaptation methods and argue that steering should be viewed as a form of model adaptation.
read the paper →Steering vectors can be combined in a linear fashion, as long as they are sufficiently orthogonal to each other and injected at the appropriate layers. We experiment with composing three attributes: language identity, jailbreak behavior, and conciseness. (!) The first step towards simple compositional steering.
paper coming soonA working theory: models extract language-specific features and identify the input language, then reason in a shared cross-lingual space, then differentiate the output back into the target language. (!) The ultimate, somewhat ambitious, goal of the PhD.
no dedicated paper yet