Using Comparable Corpora to Improve the Effectiveness of Cross-Language Information Retrieval

Abstract

AbstractLarge-scale comparable corpora became more abundant and accessible than parallel corpora, with the explosive growth of the World Wide Web. From the Cross-Language Information Retrieval point of view, limitation of translation resources as well as ambiguity arising due to failure to translate query terms is largely responsible for large drops in the effectiveness below monolingual performance. Therefore, strategies on bilingual terminology extraction from comparable texts must be given more attention in order to enrich existing bilingual lexicons and thesauri and to enhance Cross-Language Information Retrieval. In the present paper, we focus on the enhancement of Cross-Language Information Retrieval using a two-stage corpus-based translation model that includes bi-directional extraction of bilingual terminology from comparable corpora and selection of best translation alternatives on the basis of their morphological knowledge. The impact of comparable corpora on the performance of the Cross-Language Information Retrieval process is evaluated in this study and the results indicate that the effect is clearly positive, especially when using the linear combination with bilingual dictionaries and Japanese-English pair of languages.KeywordsAverage PrecisionQuery ExpansionSource LanguageTranslation ModelParallel CorpusThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.