Cross Language Information Retrieval for Biomedical Literature

This workshop report discusses the collaborative work of UT, EMC and TNO on the TREC Genomics Track 2007. The biomedical information retrieval task is approached using cross language methods, in which biomedical concept detection is combined with effective IR based on unigram language models. Furthermore, a co-occurrence method is used to select and filter candidate answers. On its own, the cross lingual approach and the filtering do not strongly improve retrieval results. However, the combination of approaches does show a strong improvement over the monolingual baseline.