CEF eTranslation tool among the best in the field

At the close of the Fifth Conference on Machine Translation (WMT20) on 19 and 20 November 2020, the Connecting Europe Facility's (CEF) machine translation tool, eTranslation, proved itself to be among the very best in the world of machine translation and computational linguistics.

This annual conference gathers the biggest players in the industry, who then complete a set of common translation tasks (such as building translation engines specifically for the news sector), sharing how they completed them and comparing the results. It is a great opportunity for the CEF eTranslation team to benchmark their achievements and capabilities against others at the cutting edge of machine translation technology. In this spirit of healthy competition, participants not only get to find out their relative standing in the field, but also to share their knowledge, experience and best practices with others. 

Register to eTranslation

eTranslation highlights

The CEF eTranslation team completed the task of building translation engines specifically for the news sector across five languages. They translated from English to the only three EU languages available at this year's Conference, Czech, German and Polish, and also translated from Japanese and Russian into English. These two important world languages are just the start of eTranslation's commitment to expand its portfolio beyond its core of official EU languages, which already includes Mandarin and Turkish.

Translations were ranked both automatically (by BLEU score) and by human judges. BLEU showed a range of translation quality, with eTranslation always in the top tier. When the human eye came into play, eTranslation again fared well, and was typically in the top-ranked "cluster" (grouping of engines only separated by expected margins of error).

Notably, eTranslation ranked 2nd for translations into Polish, outperforming all four other publicly available online translation systems participating. 

The conference also requires comparisons with human translations to see how much machine translation is closing the gap. While human translation remains the gold standard, for Russian to English translation, eTranslation was one of four engines that gave better results than the human translation used for the comparison. 

WMT20 also tests for the "robustness" of MT systems: how well can they translate texts with a different style and/or content than those they were built for. Again, eTranslation placed well, coming third out of nine for English to German, and in the second of three "clusters" for Japanese to English.

The future of translation

The results of this conference clearly show that the quality of eTranslation's machine translation engines is more or less on a par with the very best in the field. While parity with human translations has not been reached, the steady improvement in the field means that machine translation is having an ever greater impact in the field of translation, both as a tool for human translators and as an alternative where certified accuracy and perfect style are not required.

The eTranslation team will be closely examining the choices made by other participants in their engines for useful tips and tricks to maintain their current success and stay at the forefront of machine translation technology.