Learning Bilingual Translations from Comparable Corpora toCross-Language Information Retrieval: Hybrid Statistics-based andLinguistics-based Approach

摘要

Recent years saw an increased interestrnin the use and the construction of largerncorpora. With this increased interestrnand awareness has come an expansionrnin the application to knowledge acquisitionrnand bilingual terminology extraction.rnThe present paper will seek tornpresent an approach to bilingual lexiconrnextraction from non-aligned comparablerncorpora, combination to linguisticsbasedrnpruning and evaluations on Cross-rnLanguage Information Retrieval. We proposernand explore a two-stages translationrnmodel for the acquisition of bilingual terminologyrnfrom comparable corpora, disambiguationrnand selection of best translationrnalternatives on the basis of their morphologicalrnknowledge. Evaluations usingrna large-scale test collection on Japanese-rnEnglish and different weighting schemesrnof SMART retrieval system confirmed therneffectiveness of the proposed combinationrnof two-stages comparable corporarnand linguistics-based pruning on Cross-rnLanguage Information Retrieval.