The competition, called 'General MT Shared Task,' was held as part of the Workshop for Machine Translation 2022 (Conference on Machine Translation), which is the most prestigious conference in the field of machine translation.

 

“Great success of Poles. PhD candidates and students of Mathematics and Computer Science at the Adam Mickiewicz University proved better than Google and Microsoft's translator! A. Nowakowski's team won the world's most important automatic translation competition with their neural networks. They will receive their prize in Abu Dhabi,” the President of Lem Institute Maciej Kawecki wrote on Twitter. 

 

The researchers' task was to create the best possible translator, able to translate texts from different fields. 

 

“This year, the main fields the competition focused on were news, social media, human conversation (e.g., chat) and e-commerce,” Artur Nowakowski explains, who led the team. 

 

“We took part in the Czech-Ukrainian and Ukrainian-Czech translation directions. According to the published results based on automatic evaluation (where the machine assesses the quality of the translation), we came first in both directions," adds the scientist. 

 

It is worth mentioning that the team only used the data provided by the competition organisers, unlike systems that could use any resources, which usually lead to much better translation quality, such as Google or Microsoft systems.

 

The creation of the translation system took about two months. “During this time, we tried various methods based on neural networks that would be able to produce the best possible translation. In order to create the system, we needed a powerful computing server, which we purchased from the UAM's Centre for Artificial Intelligence,” says Artur Nowakowski. Both he and Gabriela Pałka are collaborating with CSI.

 

The team consisted of PhD candidates Artur Nowakowski and Gabriela Pałka and two master's students Mikołaj Pokrywka and Kamil Guttmann.

 

A conference summarising this year's results will be held in Abu Dhabi (United Arab Emirates) on 7-8 December. By then, all participating translation systems will also undergo a human evaluation, where professional translators will manually assess the quality of the translations generated by the system.