Abstract
This paper describes an architecture of a machine translation system designed primarily for Slavic languages. The architecture is based upon a shallow transfer module and a stochastic ranker.
The shallow transfer module helps to resolve the problems, which arise even in the translation of related languages, the stochastic ranker then chooses the best translation out of a set provided by a shallow transfer. The results of the evaluation support the claim that both modules newly introduced into the system result in an improvement of the translation quality.