Research to improve Cross-Language Retrieval. Position Paper for CLEF

Abstract. Improvement in cross-language information retrieval results can come from a variety of sources { failure analysis, resource enrichment in terms of stemming and parallel and comparable corpora, use of pivot languages, as well as phonetic transliteration and Romanization. Appli-cation of these methodologies should contribute to a gradual increase in the ability of search software to cross the language barrier. 1 Failure Analysis In my opinion there has been a dearth of detailed failure analysis in cross-language information retrieval, even among the best-performing methods in com-parative evaluations at TREC, CLEF, and NTCIR[7]. Just as a post-mortem can determine causation in mortality, a query-by-query analysis can often shed light on why some approaches succeed and others fail. Among the sets of queries utilized in these evaluations we always nd several queries which all partici-pants perform poorly (as measured by the median precision over all runs for that query). When the best performance is signicantly better than the media