Corpus-Based Query Translation for Translingual Information Retrieval

Translingual IR (TIR) -- also called Cross-Language IR -- has recently begun to receive considerable attention, primarily driven by the increased availability of international text collections, including the World Wide Web. The most popular approaches in TIR use a machinereadable dictionary (MRD) for query translation; their performance relative to bilingual-corpus based approaches has not previously been evaluated. In a recent set of experiments, both machine-translation and IR approaches to TIR were compared, including the corpusbased query translation and the general-purpose bilingual dictionary-based query translation described here. Since most real-life queries consist of isolated words or very short phrases, full-fledged machine translation is not applicable to query translation. Instead, various word-based approaches were investigated..