Principal investigator (ÚFAL): 
Grant id: 
ÚFAL budget: 
CZK 7.1M


Natural multi-domain dialogue systems

This project establishes a new research group for dialogue systems (natural-language human-machine communication). This field is booming and the university lacks the related research, teaching, and student involvement.

Research-wise, the project transforms the state-of-the-art in the field by extending the range of possible human-machine dialogues and making them more natural. The use of statistical methods increases system flexibility and reduces development costs.

The main aim of the project is to create a dialogue system capable of natural dialogue in multiple communication domains, including social chit-chat. This system unifies and improves the functions of voice assistants and chatbots. Previously, the former would only react to preset commands, the latter would simulate chit-chat without truly understanding. The system is based on modern machine learning methods, particularly deep neural networks.

The system created in the project will also be deployed for Czech, where the availability of dialogue systems is limited due to high costs of adapting handcrafted rules. The project further aims to develop new, effective methods for automatic dialogue system evaluation.



  • Ondřej Dušek, Jekaterina Novikova, and Verena Rieser. Evaluating the State-of-the-Art of End-to-End Natural Language Generation: The E2E NLG Challenge, In: Computer Speech and Language. [ScienceDirect / arXiv / web]
  • Ondřej Dušek, Karin Sevegnani, Ioannis Konstas, and Verena Rieser. Automatic Quality Estimation for Natural Language Generation: Ranting (Jointly Rating and Ranking), In: INLG, Tokyo. [arXiv / slides / Github]
  • Ondřej Dušek, David M. Howcroft, and Verena Rieser. Semantic Noise Matters for Neural Natural Language Generation, In: INLG, Tokyo. [PDF / poster / Github]
  • Ondřej Dušek and Filip Jurčíček. Neural Generation for Czech: Data and Baselines, In: INLG, Tokyo. [arXiv / slides / Github (code) / Github (data)]
  • Simon Keizer, Ondřej Dušek, Xingkun Liu, and Verena Rieser. User Evaluation of a Multi-dimensional Statistical Dialogue System, In: SIGDIAL, Stockholm.  [ACL / arXiv / poster / code]


New courses developed in the project, teaching started in 2019:

  • Dialogue systems – Summer semester. Bachelor's study introductory course into dialogue systems.
  • Statistical dialogue systems – Fall semester. Master's study advanced course, more emphasis on machine learning and neural networks.