Tilde machine translation technology announced the world’s most accurate for the 3rd time, outrunning Google, Microsoft and others

  • 2019-08-06
  • TBT Staff

For the third consecutive year, the Baltic-based language technology company Tilde has won the “Olympics” of the machine translation technology – the WMT 2019 competition. This year Tilde team participated with its machine translation technology for the English-Lithuanian language pair, outperforming online translation platforms such as Google, Microsoft, and others.

“The continuous success at WMT proves that we have all the necessary skills and resources to develop outstanding artificial intelligence technologies, specifically by focusing on difficult languages and complex linguistic aspects. This achievement is the result of our long-term investment in research and close cooperation with the industry-leading European universities and research centers. 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.

The WMT competition has already been taking place for 13 years, and the competing teams include large tech companies, research centers, and universities from all around the globe. Since 2017 the competition has included tasks in the Baltic languages.

Last year, Tilde won the WMT competition with the Estonian-English machine translation tool, and in 2017 - was announced as the best in the Latvian-English language pair translation task. This year, Tilde team participated with its latest neural machine translation system for the Lithuanian language and took the prize for the third time in a fierce competition with the leading technology companies and research centers in the world. 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 a professional 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 who built the system in 4 months.

Machine translation technology helps to take down language barriers in communication, cut operational costs, and increase the productivity of translators, journalists, and other professionals. Tilde is developing customized machine translation solutions for public institutions and businesses with the specific terminology and communication language style that offer such benefits as more efficient information circulation and reduction of translation costs by more than 35%.

The latest neural machine translation systems have helped Tilde make a mark as the leading machine translation technology service provider. Already for three years, Tilde has been providing the presidency of the Council of the EU with its customized machine translation tool which assists the organizers, translators, guests and media representatives to translate various texts, documents and web pages in the majority of the official EU languages. This helps people overcome language barriers and ensures that all of the information is accessible in the official language of each member state.

Tilde machine translation technologies are freely accessible for Latvian, Lithuanian, Estonian, Finnish, English, Russian, Polish and Arabic languages here: translate.tilde.com. It allows one to translate not only plain text, but also larger documents and web pages.

About Tilde

As a leading European language technology company, Tilde drives innovation in AI-based text and speech processing for complex languages. With advanced localization services, smart virtual assistants, custom machine translation systems, and online terminology tools, Tilde eliminates language barriers in global communication. Tilde’s innovative solutions are used by multinational corporations, EU public administrations, major language industry companies, as well as thousands of SMEs and individual users.