The Bridging AI and Neuroscience (BrAIN) group at the Max Planck Institute for Software Systems (MPI-SWS) in Saarbrücken, Germany, led by Dr. Mariya Toneva, is seeking an outstanding Postdoctoral Researcher in Natural Language Processing and Human Behavioral Evaluation to join the newly funded ERC Starting Grant project BrainAlign.

About the project

BrainAlign aims to develop a new generation of language models whose internal processing is aligned with the human brain. The project combines machine learning, natural language processing, and cognitive neuroscience to investigate whether brain alignment can give language models more human-like understanding of natural language, particularly their ability to integrate information over long timescales.

A major goal of the project is to go beyond existing evaluations of long-context language models. Rather than relying primarily on short excerpts or synthetic retrieval tasks, BrainAlign will develop new ways of evaluating language understanding over entire naturalistic narratives and books. The project will combine large-scale human behavioral data with controlled computational experiments to determine which linguistic capabilities are improved by brain alignment.

The project is funded by the European Research Council (ERC) and will involve a team of three postdocs and two PhD students working closely with the PI and collaborators across AI and neuroscience.

Your role

You will lead the human behavioral evaluation and long-range language understanding component of BrainAlign. You will develop a novel gamified online platform through which readers can answer questions about books and other long-form narratives. This platform will enable the collection of large-scale behavioral data about natural language understanding across entire books, creating new datasets for evaluating both humans and language models.

In particular, you will:

• Design and develop behavioral tasks for long-range language understanding based on complete natural-language narratives and books.
• Develop a gamified platform for collecting large-scale behavioral data from readers.
• Analyze human behavioral data.
• Develop and apply evaluation methods for assessing the linguistic capabilities of brain-aligned language models (BALMs).
• Collaborate closely with the other BrainAlign researchers developing the underlying brain-aligned models and contribute to joint publications in leading NLP, ML, and cognitive science venues.
• Publish your research in leading venues in machine learning, natural language processing, and computational/cognitive neuroscience.

A particularly exciting aspect of this position is the opportunity to develop new scientific paradigms for evaluating language understanding rather than simply applying existing benchmarks. The project asks fundamental questions about what it means for an AI system to understand language over the same timescales that humans naturally do.

Who we are looking for

We are looking for candidates with a strong background in natural language processing, machine learning, cognitive science, psycholinguistics, or behavioral research. You should have:

• A PhD (or expected PhD completion before the start date) in natural language processing, computer science, cognitive science, psychology, psycholinguistics, or a related discipline.
• Strong experience with NLP and/or evaluation of language models.
• Experience designing and analyzing human behavioral experiments is highly desirable.
• Strong quantitative and programming skills, preferably in Python.
• Experience with experimental design, statistical analysis, or crowdsourcing/online behavioral data collection is a plus.
• A strong publication record relative to career stage.
• The ability to work independently and across disciplinary boundaries.
• Excellent written and spoken English.

Experience with large language models, long-context language understanding, psycholinguistics, narrative comprehension, human-centered AI, or online experimentation would be particularly valuable.

What we offer

You will join an ambitious research program at the intersection of natural language processing, AI, and cognitive neuroscience, with substantial freedom to develop your own research directions within the broader goals of BrainAlign.

The position offers the opportunity to work on a genuinely new research direction: using large-scale human behavioral data and brain-aligned AI to understand and improve long-range language comprehension. You will work closely with researchers developing the computational and neuroimaging components of the project, providing an unusually integrated perspective across AI and human language understanding.

MPI-SWS offers an international and highly collaborative research environment in Saarbrücken, one of Europe's major centers for computer science and AI research. The position includes opportunities for international conference travel and research collaborations.