Hi there!

A picture of myself

I’m Lukas, a postdoc in the LEADS group led by Limor Raviv at the Max Planck Institute for Psycholinguistics. I’m most passionate about natural language processing and continual learning. Right now, I’m focusing on how deep nets learn to communicate and how that relates to human language learning.

I did my PhD on text and graph representation learning with Ansgar Scherp at Kiel University, where I also obtained my Master’s degree in Computer Science. I’ve worked on text classification, information retrieval, self-supervised text representation learning, multimodal autoencoders, and lifelong graph learning.

Besides research, I enjoy surfing, vanlife, and other sorts of outdoor adventures. I also like playing chess, doing yoga, and making music. I choose vim over emacs and actively prevent myself from spending more time with vimscript than the result would save me.

Selected publications

  • Galke, L., Franke, B., Zielke, T., & Scherp, A. (2021). Lifelong Learning of Graph Neural Networks for Open-World Node Classification. 2021 International Joint Conference on Neural Networks (IJCNN), 1–8. [paper] [code]
  • Galke, L., & Scherp, A. (2022). Bag-of-Words vs. Graph vs. Sequence in Text Classification: Questioning the Necessity of Text-Graphs and the Surprising Strength of a Wide MLP. Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 4038–4051. [paper] [code]
  • Vagliano, I., Galke, L., & Scherp, A. (2022). Recommendations for item set completion: On the semantics of item co-occurrence with data sparsity, input size, and input modalities. Information Retrieval Journal. [paper] [code]

To see more of my publications, visit my Google Scholar profile or my DBLP profile.

Current project

Neural networks are still far behind human capabilities in generalizing from few examples and continual learning. I want to fix that.

My current research investigates how neural network agents learn to communicate. Imagine you put some (artificial, of course) neural network agents into a game that can only be solved with communication. How do communication protocols emerge? How similar are those to emergent languages of humans in the same setting? What can we learn from humans to improve our models?

This relates very well to the field of emergent communication. My first baby-step into the field is a workshop paper that summarizes recent progress of the emergent communication field and contrasts it with linguistic phenomena in humans. Very recently, I co-organized a workshop on machine learning for language evolution research at the Joint Conference for Language Evolution.

Past projects


I’m usually open to collaborate as long as it’s mildly related to what I’m doing. To get in touch, just drop me an e-mail via lukas.galke@mpi.nl. Other than that, feel free to reach out via Mastodon.