A comparative evaluation of three machine translation systems  – Tilde, Google Translate, and Microsoft’s Bing Translator – has confirmed that Tilde produces higher-quality translations for Baltic languages. The evaluation was performed using the BLEU (Bilingual Evaluation Understudy) score metric, which evaluates how closely a machine-translated text compares with a professional human-translated document.


In the evaluation, Tilde performed better than Google Translate and Microsoft’s Bing Translator for the following language combinations: English-Latvian, Latvian-English, English-Lithuanian, Lithuanian-English, English-Estonian, and Estonian-English. In each instance, the BLEU score was considerably higher for Tilde (see the chart below).


Tilde’s machine translation technology LetsMT was developed with support from the European Commission, along with a consortium of researchers from leading universities in Europe. Tilde can customize MT systems for any language combination, and the technology is ideally suited for smaller languages. Customized systems are trained with terminology sets, translation memories, and glossaries.

References

Vasiļjevs, A., Skadiņš, R., & Tiedemann, J. (2012). LetsMT!: Cloud-Based Platform for Do-It-Yourself Machine Translation. In Proceedings of the ACL 2012 System Demonstrations (pp. 43–48). Jeju Island, Korea: Association for Computational Linguistics. Retrieved from http://www.aclweb.org/anthology/P12-3008.

Skadiņš, R., Puriņš, M., Skadiņa, I., & Vasiļjevs, A. (2011). Evaluation of SMT in localization to under-resourced inflected language. In M. L. Forcada, H. Depraetere, & V. Vandeghinste (Eds.), Proceedings of the 15th International Conference of the European Association for Machine Translation EAMT 2011 (pp. 35–40). Leuven. Retrieved from http://www.mt-archive.info/EAMT-2011-Skadins.pdf.

Skadiņš, R., Goba, K., & Šics, V. (2010). Improving SMT for Baltic Languages with Factored Models. In Proceedings of the Fourth International Conference Baltic HLT 2010, Frontiers in Artificial Intelligence and Applications, Vol. 2192 (pp. 125–132). Riga: IOS Press.