In this paper we extend previous work on document retrieval across multilingual corpora. In this setting it is often assumed that we have a certain alignment given based on which we can learn mapping
