Translingual Information Retrieval: A Comparative Evaluation

Abstract

Translingual information retrieval (TIR) consists of providing a query in one language and searchin g document collections in one or mor e different languages. This paper introduc es n ew TIR methods and reports on comparative TIR expe rim ents with these new methods and with previously reported ones in a realistic setting. Methods fall into two categories: query transl at ion based , and statistical-IR approaches establishing translingua l associations. The results show that using bilingual corpora for automated ext raction of term equivalences in context outperforms other m et hods. Translingual versions of the Generalized Vector Space Model (GVSM) an d Lat ent Semantic Indexing (LS I) perform relatively well, as does translingua l pseudo relevance feedback (PRF). All showed re lativ ely small performance loss between monolingual and translingual versions. Query translation based on a general machinereadable bilingual dictionary - heretofore the most popular m et hod - did not mat ch the performance of other, more sophisticated m ethod s. Also , the previous very high LSI r es ults in the literatur e were disconfirmed by more r ea listi c releva nc e- based evaluations.