Saarland University is a campus university that is internationally recognized for its strong research programmes.
Fostering young academic talent and creating ideal conditions for teaching and research are a core part of the university’s mission.
As part of the University of the Greater Region, Saarland University enables students and staff to share and exchange knowledge and ideas between disciplines, between universities and across borders.
With over 17,000 national and international students, studying more than a hundred different academic disciplines, Saarland University is a diverse and dynamic learning environment.
Saarland University is officially recognized as one of Germany’s family-friendly higher-education institutions and with a combined workforce of more than 4,000 it is one of the largest employers in the region.
The research groups of Michael Hahn and Alexander Koller are inviting applications for the following position commencing

at the earliest opportunity.
Academic research assistant (m/f/x) Reference number W2846
, salary in accordance with the German TV-L salary scale 1 , pay grade:
E13 TVL, duration of employment:
3 years, volume of employment:
100 % of standard working time.

Workplace/Department:

The position is funded through the new project "Structural generalization in transformer- based LLMs", which is part of the DFG Special Priority Program Robust Assessment and Safe Applicability of Language Modeling:
Foundations for a New Field of Language Science and Technology.
The goal of the SPP is to bring together research in linguistics and LLMs so they can mutually inform each other.
The starting point of the project is the observation that transformers struggle with *structural generalization:* they do not perform well on test instances that are more complex than the training instances.
We see this e.g. when parsing complex sentences or when managing complex reasoning chains.
In the project, we want to develop theory that explains this limitation, carry out empirical research to pinpoint the transformer’s capabilities, understand them through mechanistic interpretability, and find ways to improve structural generalization in transformers.
Methods range from formal language theory to training and promptingbased experiments to circuit analysis.
We will carry out the project in collaboration with Will Merrill and Yuekun Yao.

1 TV-L = collective agreement on remuneration of public sector employees in the German Länder

Job requirements and responsibilities:

We are looking for candidates who have finished, or are about to complete, an excellent Master’s degree in computer science, computational linguistics, or a related discipline.
The ideal candidate will have outstanding programming skills and algorithmic understanding; a strong understanding of current methods in machine learning; and strong communication skills in English (spoken and written).

What we can offer you:

o A flexible work schedule allowing you to balance work and family, among other things the possibility of teleworking o Secure and future-oriented employment with attractive conditions o A broad range of further education and professional development programmes (for example language courses) o An occupational health management model with numerous attractive options, such as our university sports programme o Supplementary pension scheme (RZVK) o Discounted tickets on local public transport services (’Job-Ticket’ of the saarVV) o Job bike leasing (JobRad) We look forward to receiving your

meaningful online application
(in a PDF file) by