2014

2015

2016

2017

2018

2019

2020

2022

2023

2024

The 60th meeting of the Prague computer science seminar

Ondřej Bojar

AI Stolen by Transformers!

Sequence-to-sequence deep learning models originating in the area of machine translation (MT) have exploded the public interest in AI, effectively stealing the name of the field for the current years. To MT researchers, this is reminiscent of 2013-2015 when deep learning invaded the field of MT and completely rewrote the methodology.

Annotation Lecturer Poster

2024
2023
2022
2020
2019
2018
2017
2016
2015
2014

Play record

May 23, 2024

4:15pm

Auditorium E-107, FEL CTU
Karlovo nám. 13, Praha 2
Show on the map

In my talk, I will mention some of our achievements in translation thanks to Transformers and demonstrate our heavily multilingual speech-to-text translation, but I will primarily illustrate and warn about common misconceptions and evaluation fallacies we know well from the MT domain, e.g. attributing observed gains to wrong reasons.

Unfortunately, the current AI hype is fuelled to some extent by such mismeasurements. If we do not bring more technically sound and realistic assessment of large language models' abilities into the discussion, we are risking another AI winter, i.e. a sudden decline in interest and both private and public funding into AI development.

Ondřej Bojar

Ondřej Bojar is an Associate Professor at ÚFAL (Institute of Formal and Applied Linguistics), Charles University, and a lead scientist in Machine Translation in the Czech Republic. He has been co-organizing a well-known series of shared tasks in machine translation and machine translation evaluation (WMT) since 2013. His system has dominated English-Czech translation in the years 2013-2015, before deep learning and neural networks fundamentally changed the field. Having taken part and later supervised ÚFAL’s participation in a series of EU projects (EuroMatrix, EuroMatrixPlus, MosesCore, QT21, HimL, CRACKER, Bergamot), he has recently concluded his coordination of the EU project ELITR focused on simultaneous speech translation into over 40 languages. ELITR has also coined the task of project meeting summarization with its AutoMin 2021 and 2023 shared task.

Its program consists of a one-hour lecture followed by a discussion. The lecture is based on an (internationally) exceptional or remarkable achievement of the lecturer, presented in a way which is comprehensible and interesting to a broad computer science community. The lectures are in English.

The seminar is organized by the organizational committee consisting of Roman Barták (Charles University, Faculty of Mathematics and Physics), Jaroslav Hlinka (Czech Academy of Sciences, Computer Science Institute), Michal Chytil, Pavel Kordík (CTU in Prague, Faculty of Information Technologies), Michal Koucký (Charles University, Faculty of Mathematics and Physics), Jan Kybic (CTU in Prague, Faculty of Electrical Engineering), Michal Pěchouček (CTU in Prague, Faculty of Electrical Engineering), Jiří Sgall (Charles University, Faculty of Mathematics and Physics), Vojtěch Svátek (University of Economics, Faculty of Informatics and Statistics), Michal Šorel (Czech Academy of Sciences, Institute of Information Theory and Automation), Tomáš Werner (CTU in Prague, Faculty of Electrical Engineering), and Filip Železný (CTU in Prague, Faculty of Electrical Engineering)

The idea to organize this seminar emerged in discussions of the representatives of several research institutes on how to avoid the undesired fragmentation of the Czech computer science community.