Issues in Pre- and Post-translation Document Expansion:Untranslatable Cognates and Missegmented Words

摘要

Query expansion by pseudo-relevancernfeedback is a well-established techniquernin both mono- and cross- lingual informationrnretrieval, enriching and disambiguatingrnthe typically terse queries providedrnby searchers. Comparable document-sidernexpansion is a relatively more recent developmentrnmotivated by error-prone transcriptionrnand translation processes in spokenrndocument and cross-language retrieval.rnIn the cross-language case, onerncan perform expansion before translation,rnafter translation, and at both points. Werninvestigate the relative impact of pre- andrnpost- translation document expansion forrncross-language spoken document retrievalrnin Mandarin Chinese. We find that posttranslationrnexpansion yields a highly significantrnimprovement in retrieval effectiveness,rnwhile improvements due to pretranslationrnexpansion alone or in combinationrndo not reach significance. We identifyrntwo key factors of segmentation and translationrnin Chinese orthography that limitrnthe effectiveness of pre-translation expansionrnin the Chinese-English case, whilernpost-translation expansion yields its fullrnbenefit.