Cross Language Information Retrieval Using Parallel Corpus with Bilingual Mapping Method

Abstract

This study presents an approach to generate a bilingual language model that will be used for CLIR task. Language models for Bahasa Indonesia and English are created by utilizing a bilingual parallel corpus, and then the bilingual language model is created by learning the mapping between the Indonesian model and the English model using the Multilayer Perceptron model. Query expansion is also used in this system to boost the results of the retrieval, using pre-Bilingual Mapping, post-Bilingual Mapping and hybrid approaches. The results of the experiments show that the implemented system, with the addition of pre-Bilingual Mapping query expansion, manages to improve the performance of the CLIR task.