The lack of parallel corpora and linguistic resources for many languages and domains is one of the major obstacles for the further advancement of automated translation. A possible solution is to exploit comparable corpora (non-parallel bi- or multi-lingual text resources) which are much more widely available than parallel translation data.