The goal of Prof. Milica Gasic' DFG project "Curriculum learning in natural language generation” is to develop a deeper understanding of how and why hallucinations arise in neural natural language generation models. Rather than treating hallucinations solely as an undesirable property of the final output, we investigate them as a phenomenon emerging from the interaction between data, learning objectives, representations, and the mechanisms by which models acquire and use knowledge. A central goal is to identify the conditions under which hallucinations emerge and, ultimately, develop principled approaches for reducing their magnitude and severity. The project will investigate learning mechanisms underlying reliable language generation, with particular emphasis on curriculum learning and learning from feedback, exploring approaches such as inverse reinforcement learning and their connection to Theory of Mind.

The candidate must hold or be close to obtaining a master’s degree in Computer Science, Mathematics, Engineering, Computational Linguistics or a related field. The master’s thesis must have been submitted by the time of application. Some exposure to Natural Language Processing or Machine Learning is desirable. Excellent programming skills are essential. The candidate should be fluent in English with good writing skills. Knowledge of German is not required, but a willingness to learn German is desirable.

The position comes with a salary representative in academia (EG 13 TV-L, starting at € 4,759.37; classification depends on prior professional experience). The position is initially a one-year fixed-term contract, with the option of extension based on satisfactory performance. There will be no teaching duties in the first year. It is a qualification position in the sense of the Act of Academic Fixed-Term Contract (Wissenschaftsvertragsgesetz-WissZeitVG), which is to promote the scientific qualification of the employee. The start date is negotiable, but the preferred start would be on 1st January 2027.

Company pension grant; Good transport connection; Free parking; Working from home; International work; Bonus payments; Union rates; Payment of capital-forming benefits

The Dialogue Systems and Machine Learning group at Heinrich Heine University investigates fundamental questions in human–computer interaction, at the intersection of Natural Language Processing and Machine Learning. Our research focuses on building intelligent dialogue systems that can learn, reason, adapt, and interact effectively with humans in dynamic and uncertain environments. We develop principled methods for dialogue modelling, representation learning, reinforcement learning, knowledge acquisition, and user simulation. Our mission is to advance the foundations of machine intelligence towards dialogue systems that are adaptive, trustworthy, and genuinely useful to people.

Frau Prof. Milica Gasic
gasic@hhu.de