SPIDER Retrieval System at TREC7

This year the Zurich team participated in two tracks: the automatic-adhoc track and the crosslingual track. For the adhoc task we focused on improving retrieval for short queries. We pursued two aims. First, we investigated weighting functions for short queries|explicitely without any kind of automatic query expansion. Second we developed rules that automatically decide for which queries automatic expansion works ne and for which it does not. For the cross-language track, we approached the problem of retrieving documents from a multilingual document pool containing documents in all TREC CLIR languages. Our method uses individual runs for di erent language combinations, followed by merging their results into one nal ranked list. We obtained good results without sophisticated machine translation or costly linguistic resources.