Rudolf Rosa

office hours
Mon 15:00-19:00
Tue 15:30-19:00
These are my usual office hours when you can find me with a high probability.
Feel free to e-mail me for an appointment on other days/at other times!
Malostranské náměstí 25
118 00 Praha 1
Czech Republic

Main Research Interests

  • Looking for linguistic structures in Deep Neural Networks
  • Morphology, derivations
  • Dependency parsing
  • Unsupervised and semi-supervised methods, especially cross-lingual and multilingual
  • In the past: automatic post-editing of Machine translation


Linguistic Structure Representation in Neural Networks (LSD)

I am now working on a GAČR grant of David Mareček, called LSD, where we are trying to look at what linguistic structures can be found hidden inside of neural networks.

Unsupervised morphology induction

Together with Zdeněk Žabokrtský, we are trying to handle morphology in an unsupervised way, e.g. to find lemmas for word forms, to separate derivation from inflection, etc.


  • Cross-lingual Syntactic Parsing, i.e. training a parser on one language and applying it to another language. This was my dissertation, and I also had a GAUK grant for that.
  • Pohádkové dítě / Fairytale Child chatbot. A simple console chatbot that wants to hear a fairly tale from you! / Jednoduchý konzolový chatbot, který si od vás chce nechat vyprávět pohádku!
  • HimL focused on semantically sane translation of medical texts from English to Czech, German, Romanian and Polish.
  • QTLeap was a project aimed at significantly improving the quality of machine translation using deep language processing approaches (also see TectoMT).
  • I was a member of the HamleDT group, which was a project of harmonizing dependency treebanks for various languages, later evolving and merging into the Universal Dependencies project (which I am also an official memebr of).
  • Depfix is a system for automatic post-editing of machine translation outputs. It was developed as a part of the Faust project. It was later succeeded by MLFix, by Dušan Variš.
  • MSTperl is a reimplementation of the Maximum spanning tree dependency parser (McDonald et al., 2005) in Perl. It is tuned for Czech and has several advanced features that are useful for parsing the machine-translated sentences by Depfix. It also has some features for delexicalized parser transfer. I do not use it anymore -- I switched to Parsito and UDPipe.

Curriculum Vitae

You can download my CV either in English or in Czech.


List of classesNPFL092 NLP TechnologyNPFL118 Natural language processing on computational clusterNPFL120 Multilingual Natural Language ProcessingNPRG045


Stránka cvičení Programování 1

I am happy to supervise NLP projects (bachelor theses, master theses, etc.), have a look at Project Ideas.

Rád povedu projekty v oblasti zpracování přirozeného jazyka (Bc. a Mgr. práce apod.), mrkněte na Náměty na projekty.

Selected Bibliography


Bachelor students

  • Yuliya Yamalutdinova: Contradiction Detection


I am one of the main organizers of the Slovakoczech NLP workshop for students and early-stage researchers -- see SloNLP 2015, SloNLP 2016SloNLP 2017SloNLP 2018.

I am the student ambassador of ÚFAL, so feel free to contact me with anything related to studying here. / Jsem studentským ambasadorem ÚFALu -- zeptejte se mě na cokoliv ohledně studia lingvistiky na Matfyzu!

My ORCID ID is 0000-0003-4908-6127.

My Erdös number is 4 (me - Jaroslava Hlaváčová - Petr Savický - Zsolt Tuza - Paul Erdös)

Můj herní index je 38 (aktuální před Tmou 2016), náš šifrovací tým se jmenuje Divize nulou.