Other Related Approaches

Having a bilingual document-aligned training corpus makes it possible to use adaptations of other IR approaches:

• GVSM (generalized vector space model): Represent terms in either language by their pattern of occurrence in the training corpus.
• Pseudo-relevance feedback: Use query in language X to retrieve documents in language X. Replace returned documents by their equivalents in language Y. Use the top documents in language Y as a query against the full document collection in language Y.

Paper describing these ideas by a group at CMU's language technology institute.