University of Tartu took Estonian language machine translation to technology business

Language technology researchers at UT Institute of Computer Science started cooperation in machine translation with translation agency Grata. The aim of the cooperation is to create a machine translation solution for automatic translation from German and English into Estonian that takes into consideration the needs of the specific translation agency.

The application will use a statistical approach to machine translation, which means that the system learns to translate automatically, by using a huge amount of translation examples, and does not depend on the input or output language. Therefore, the solution works best with data from which it has “learned” and it also takes into account the topic of the translated texts.

“With technical specifications, user manuals or, for example, legal texts, it is possible to get a trustworthy automatic translation of a majority of the input sentences. Incorrect translations can also be identified and filtered out if it is easier to translate the original yourself,” Mark Fišel, UT Associate Professor of Language Technology, described the opportunities of machine translation. “Of course it is not automagical: for example, the statistical approach does not take into account the special features of some languages and the system output needs to be analysed and different corrections offered—this is where we are most useful.”

Using a good quality machine translation application gives the translation agency the opportunity to provide new services for customers. “By using machine translation, we hope to meet the requests of those customers who would already now like to get, for example, post-editing service, we also hope to increase productivity by integrating machine translation into the work process,” said CEO of Grata Eve Anijärv.

This is the first cooperation project for the University of Tartu where the technology of Estonian language machine translation is taken into the business world. “In rest of the world, especially in western Europe, machine translation and post-editing are actively used. Estonia is only catching up and of course we want to be at the forefront,” said Fišel.

The development work also involves students for whom it is a good opportunity to combine research work with a practical output. The Grata machine translation base system developed by language technologists should be completed in the summer of 2016.

The project is funded by the European Union Regional Development Fund.

Additional information:
Mark Fišel, UT Associate Professor of Language Technology
tel. 737 5430
e-mail: mark.fisel [ät] ut.ee

Eve Anijärv, CEO of Grata OÜ
tel. 736 6472
e-mail: eve [ät] grata.ee