HHU StartMNFComputer SciencesResearch groupsDialog Systems and Machine Learning

Dialog Systems and Machine Learning


The Dialogue Systems and Machine Learning Group conducts fundamental research in natural language processing and related areas of machine learning, with a view towards the development of the next generation of intelligent
conversational agents.  This research is currently centred around the following key problems:

  • 1. Knowledge Extraction
    Traditionally, research into dialogue systems has assumed that the knowledge with which a dialogue system operates is provided a priori.  How can we build systems that harvest their knowledge from non-structured natural language data?
  • 2. Dynamic dialogue policies
    How can we build ever-learning dialogue systems that can converse about dynamically acquired knowledge?
  • 3. User modelling
    How can we increase the  accuracy and coverage of user models in this user-centric technology?
  • 4. Reward modelling
    Can we include more nuanced measures such as intrinsic motivation, curiosity and sentiment to make dialogue systems more human-like?

Topology of Word Embeddings: Singularities Reflect Polysemy

Our paper “Topology of Word Embeddings: Singularities Reflect Polysemy” has been selected to receive the Best Paper Award at *SEM 2020.

Post-doctoral Position in Dialog Modelling

We are looking for an enthusiastic and talented post-doctoral researcher to join our award-winning international research team at Heinrich Heine University Düsseldorf.

Plenary lecture "10 things you should know about dialogue”

Milica Gasic gave a plenary lecture on "10 things you should know about dialogue” at Frederick Jelinek Memorial Summer workshop on Speech and Language Technology (JSALT 2020).

Please click here to view the slides.

Office Adress

Group Dialog Systems and Machine Learning
Building 25.12 Level 01
Universitätsstraße 1
40225 Düsseldorf

No actual events


Prof. Dr. Milica Gašić

Building: 25.12
Floor/Room: 01.29
Phone +49 211 81-13787
Responsible for the content: