Computer Laboratory

Alexander Kuhnle

I am a final-year PhD student supervised by Prof Ann Copestake in the NLIP group at the Computer Laboratory, University of Cambridge, and a member of Queens' College. I am originally from Germany, where I did two bachelor degrees at the Karlsruhe Institute of Technology, the BSc in Informatics and the BSc in Mathematics, before continuing to take the MPhil in Advanced Computer Science course at the Computer Laboratory in Cambridge.

My research focuses on the evaluation of multimodal deep learning system with respect to visually grounded language understanding. Instead of using real-world data, I work on a system which automatically generates artificial toy data consisting of images of coloured geometric shapes located in a plane together with simple natural language descriptions in the style of traditional formal semantics (for instance, "Most squares are red."). Although simple, their structural complexity is still sufficient to generate a broad variety of interesting instances involving many aspects of language. In contrast to the classic image captioning or visual question answering task, the neural network here is asked to decide about the agreement of a given statement and the accompanying image. By controlling the content of training and test instances, I make sure that achieving good performance clearly indicates the acquisition of genuine language understanding and generalisation abilities (in this toy domain). Moreover, such a configurable data generator allows for more detailed investigations and comparisons of the learning process in deep neural networks and the weaknesses of specific architectures.

I am grateful for being funded by a Qualcomm Award Premium Research Studentship and an Engineering and Physical Sciences Research Council (EPSRC) Doctoral Training Studentship.

Publications and activities


  • How clever is the FiLM model, and how clever can it be?. September 2018. Alexander Kuhnle, Huiyuan Xie and Ann Copestake. Proceedings of the Workshop on Shortcomings in Vision and Language (ECCV 2018), in Munich (Germany).
  • Deep learning evaluation using deep linguistic processing. June 2018. Alexander Kuhnle and Ann Copestake. Proceedings of the Workshop on Generalization in the Age of Deep Learning (NAACL 2018), in New Orleans (USA). [pdf] [poster]
  • Artificial microworlds and deep linguistic processing for evaluating language understanding. June 2017. Alexander Kuhnle and Ann Copestake. Machine Learning Summer School (MLSS 2017), in Tübingen (Germany). [poster]
  • A proposition-based abstractive summariser. December 2016. Yimai Fang, Haoyue Zhu, Ewa Muszyńska, Alexander Kuhnle and Simone Teufel. Proceedings of the 26th International Conference on Computational Linguistics (COLING 2016), in Osaka (Japan). [pdf]
  • Evaluating multi-modal deep learning systems with micro-worlds. November 2016. Alexander Kuhnle and Ann Copestake. Cambridge Language Sciences Annual Symposium 2016, in Cambridge (UK). [abstract] [poster] [references] [slide]
  • Investigating the effect of controlled context choice in distributional semantics. August 2016. Alexander Kuhnle. ESSLLI Workshop on Distributional Semantics and Linguistic Theory (DSALT 2016), in Bolzano (Italy). [abstract] [poster]
  • Resources for building applications with Dependency Minimal Recursion Semantics. May 2016. Ann Copestake, Guy Emerson, Michael Wayne Goodman, Matic Horvat, Alexander Kuhnle and Ewa Muszyńska. Proceedings of the 10th International Conference on Language Resources and Evaluation (LREC 2016), in Portorož (Slovenia). [pdf] [github]

Working papers

  • ShapeWorld: A new test methodology for multimodal language understanding. April 2017. Alexander Kuhnle and Ann Copestake. [arxiv] [github]


  • ShapeWorld: A new testbed and evaluation methodology for multimodal language understanding. [github] [proposal]
  • [paper]
  • TensorForce: A TensorFlow library for applied reinforcement learning. Collaboration with Michael Schaarschmidt and Kai Fricke. [github] [blog]
  • Pydmrs: A Python library for working with DMRS structures. Collaboration with Ann Copestake, Guy Emerson, Matic Horvat and Ewa Muszyńska. [github]
  • [paper]

Talks etc

  • "Unit-testing" deep learning with synthetic data for more informative evaluation. June 2018. Cambridge Language Sciences Research Symposium for Early-Career Researchers, in Cambridge (UK). [slides]
  • The potential of synthetic data for more informative evaluation in Visual Question Answering. May 2018. NLIP Seminar Series, at the Computer Laboratory, University of Cambridge (UK). [slides]
  • ShapeWorld for automatic language generation in a closed-world domain. August 2017. DELPH-IN Annual Meeting, at University of Oslo (Norway). [slides]
  • "Clever" machines – Are deep neural networks able to understand language like humans?. June 2017. Grad Talks Seminar, at Queens' College, University of Cambridge (UK). [slides]
  • Natural language quantifier learning for multi-modal deep neural nets. February 2017. Research visit to the Center for Mind/Brain Sciences (CIMeC), University of Trento (Italy). Collaboration with Raffaella Bernardi, Aur\'{e}lie Herbelot, Sandro Pezzelle and Ionut Sorodoc. [report]
  • Evaluating multi-modal deep learning systems with micro-worlds. November 2016. Inaugural Postgraduate Studies Open Day, at the Computer Laboratory, University of Cambridge (UK). [slides]
  • GraphLang: A DMRS graph description language. June 2016. DELPH-IN Annual Meeting, at Stanford University (USA). Work in progress; part of the pydmrs library. [overview] [slides]



Alexander Kuhnle
Queens' College
Silver Street
Cambridge CB3 9ET
United Kingdom

aok25 (at)

Profiles:  CVLinkedInGoogle ScholarGitHubAcademiaResearchGate