Cross-language information retrieval based on weight computation of query keywords translation

In cross-language information retrieval (CLIR), the query sentence is often combined with a series of query keywords, rather than a complete natural sentence. Lack of necessary contextual syntactic information in such a query sentence makes it impossible to achieve a unique translation of the query sentence with acceptable precision. In this paper, we convert the translation of query sentence to the weight computation of the translations of the query keyword based on large-scale bilingual parallel corpora, and thereafter reconstruct the query sentence in target language. The experimental results show that the approach achieves an average retrieval accuracy of 93.4% in the front 10 retrieval results and 89.1% in the front 100 retrieval results, while the retrieval error rate is reduced by 63.62% over the purely dictionary-based baseline.