Tilde machine translation technology announced as the world's most accurate at the WMT competition

Language technologies , Machine Translation , Research
01.08.2019
Rihards Krislauks and Marcis Pinnis of Tilde MT team

Results of this year’s WMT Competition have been announced at the 4th Conference on Machine Translation in Florence, Italy on August 1-2, 2019. Tilde achieved the highest quality of machine translation for English-Lithuanian language pair, ahead of other competitors - the leading machine translation technology companies and research centers from the US, Europe, and Asia. This achievement at WMT 2019 marks the third consecutive year Tilde has taken the leader position in developing machine translation technologies for the Baltic languages.

As put by the organizers, the primary objectives of WMT are to evaluate the state of the art in machine translation, to disseminate common test sets and public training data with published performance numbers, and to refine evaluation and estimation methodologies for machine translation.

The main metric for the translation tasks is an expert judgment of translation quality for both the competing machine translation systems and the work of professional translators to be then compared with the output of most commonly used online translation platforms. According to the English-Lithuanian test results, the system developed by Tilde team received 72,8 out of 100 points, while the Microsoft team scored only 69,1. The work submitted by the human translator was rated with 90,5 points. In this competition, Tilde participated with an expert team of eight neural network architects, including four PhDs.

“The continuous success at the WMT competition proves that our research team can achieve superior translation quality by focusing on difficult languages and complex linguistic aspects. The winning technologies of Tilde can be successfully applied not only to languages of Baltic countries but other less-resourced languages and specific areas of machine translation,” comments Andrejs Vasiļjevs, Executive Chairman of Tilde.

Every discovery leading to quality improvement of the machine translation technology can significantly lower the localization costs and improve translation efficiency by more than 35% which explains the growing demand for advanced language technologies coming from companies and public institutions operating in such multilingual environments as is Europe. With the aim of preserving the European languages and the culture developed around them, EU is one of the main sources of funding the research and innovation process which, therefore, creates a fertile ground for the local language technology companies to challenge the tech giants.