How to overtake Google in MT quality - the Baltic case

Motivation of the language technology company Tilde is to improve quality of machine translation for lesser resourced languages such as the languages of Baltic countries. Generic MT solutions like Google Translate perform poorly for these complex languages. To compensate the shortage of training data and to deal with rich morphology we are applying different approaches in combining statistical methods with linguistic rules. We will present the strategies applied and the results of various experiments. We will discuss application of the production systems that show significantly better translation quality comparing to the Google Translate. We will also outline how this work contributes to creation of the European infrastructure for automated translation.