Ondřej Dušek

office
N233
office hours
email me to arrange a meeting
email
odusek@ufal.mff.cuni.cz
address
IMPAKT – „N“
V Holešovičkách 747/2
180 00 Praha 8
Czech Republic

Main Research Interests

  • Natural language generation
  • Spoken dialogue systems and chatbots
  • Evaluation methods for NLG/dialogue/NLP
  • Machine learning
  • Machine translation

Projects

Current

  • NaMuDDiS – Natural Multi-domain Dialogue Systems (Charles University funded, 2019-2021)
  • THEaiTRE – Automatically generating a theatre play (Czech Technical Agency, 2020-2022)
  • EDUbot – Education chatbot assistant (Czech Technical Agency, 2021-2023)

Past

At Heriot-Watt University in Edinburgh (2016-2018):

During my Ph.D. study (2011-2016):

  • AdaNLG – adaptive natural language generator (2014-2016)
  • Vystadial – statistical spoken dialogue system (2013-2016)
  • QTLeap – semantic machine translation (2013-2016)
  • Khresmoi – medical information retrieval (working on machine translation, 2013-2014)
  • FAUST – improving machine translation fluency (2011-2013)

Teaching

List of classesNPFL099 Statistical dialogue systemsNPFL123 Dialogue systems

Students

Students I supervise:

  • Vojtěch Hudeček (Ph.D. at Charles Uni, with Zdeněk Žabokrtský, since 2018)
  • Zdeněk Kasner (Ph.D. at Charles Uni since 2019)
  • Jonáš Kulhánek (MSc. at Charles Uni since 2020)
  • Sourabrata Mukherjee (Ph.D. at Charles Uni since 2019)
  • Tomáš Nekvinda (completed MSc. at Charles Uni, 2019–2020; Ph.D. since 2020)
  • Patrícia Schmidtová (completed BSc. at Charles Uni, with Vojtěch Hudeček, 2018–2019; MSc. since 2020)
  • Daniel Štancl (Ph.D. at Charles Uni since 2020)
  • Xinnuo Xu (Ph.D., CDT Robotics Edinburgh, with Verena Rieser & Ioannis Konstas, since 2016)
  • Borek Požár (BSc. at Charles Uni, 2020-2021)
  • Ondřej Plátek (Ph.D. at Charles Uni since 2021)
  • Jaroslav Šafář (MSc. at Charles Uni since 2021)
  • Ondřej Motlíček (MSc. at Charles Uni since 2021)
  • František Trebuňa (MSc. at Charles Uni since 2021)
  • Hana Roubalová (BSc. at Charles Uni since 2021)
  • Vojtěch John (BSc. at Charles Uni since 2021)

My former students:

News

2021/08/06: I was involved in the GEM benchmark – Generation, Evaluation & Metrics. This was presented at an ACL wokrshop and included a shared task. You can have a look at the workshop website.

2021/07/26: I have 2 papers accepted at INLG 2021 – Underreporting of errors in NLG output, and what to do about it (van Miltenburg et al., multi-party collaboration) and a shared task paper titled Text-in-Context: Token-Level Error Detection for Table-to-Text
Generation
(Kasner et al.) with Simon Mille from Pompeu Fabra University (preprint coming soon).

2021/06/15: I started my summer research visit at Prof. Milica Gašić's lab at Heinrich-Heine University Düsseldorf

2021/05/28: Our paper Shades of BLEU, Flavours of Success: The Case of MultiWOZ (Nekvinda & Dušek) was accepted to the GEM ACL Workshop.

2021/05/06: 2 papers accepted at ACL 2021 – AggGen: Ordering and Aggregating while Generating (Xu et al.) with Heriot-Watt University and Discovering Dialogue Slots with Weak Supervision (Hudeček et al.) with Zhou Yu from Columbia University.

2021/04/21: Our collaboration with LIMSI – Defining And Detecting Inconsistent System Behavior in Task-oriented Dialogues (Schaub et al.) was accepted to the TALN-RÉCITAL conference.

2021/04/26: The EDUbot project (education chatbot) funded by the Czech Technical Agency has started. We'll add more news soon.

2021/03/02: We started the 2021 run of the Dialogue Systems course.

2021/02/26: The THEaiTRE project has had the premiere of the first AI-generated play on stage!

2021/01/08: Jan Vainer won the Dean's Award for Best Master's Thesis with his work on efficient neural speech synthesis!

2020/12/17: The paper Evaluating data-to-text NLG accuracy by natural language inference by me & Zdeněk kasner has received the INLG 2020 Best Short Paper award!

2020/11/21: Two INLG 2020 papers by Zdeněk Kasner & me are now out on arXiv: Data-to-text generation by iterative text editing and Evaluating data-to-text NLG accuracy by natural language inference.

2020/11/16: The list of accepted papers and a preliminary schedule for the NLG Evaluation Workshop I'm co-organizing at INLG 2020 is now available.

2020/09/29: The Statistical Dialogue Systems course I'm teaching with Vojtěch Hudeček has started its 2020/2021 run.

2020/07/25: We got two papers on neural text-to-speech accepted to Interspeech, both are available on arXiv: Tomáš Nekvinda's One Model, Many Languages: Meta-learning for Multilingual Text-to-Speech and Jan Vainer's SpeedySpeech: Efficient Neural Speech Synthesis.

Biographical

Tools I'm participating on

  • Alex – spoken dialogue system framework
  • Flect – statistical morphology generation
  • MTMonkey – machine translation web services infrastructure
  • RatPred – trainable NLG quality estimation
  • TGen – a statistical natural language generator
  • Treex – a modular NLP toolkit

Personal webpage