Multilingual Web retrieval: An experiment on a multilingual business intelligence portal

The amount of non-English information on the Web has proliferated so rapidly in recent years that it often is difficult for a user to retrieve documents in an unfamiliar language. In this study, we report the design and evaluation of a multilingual Web portal in the business domain in English, Chinese, Japanese, Spanish, and German. Web pages relevant to the domain were collected. Search queries were translated using bilingual dictionaries, while phrasal translation and co-occurrence analysis were used for query translation disambiguation. Pivot translations were also used for language-pairs where bilingual dictionaries were not available. A user evaluation study showed that on average, multilingual performance achieved 72.99% of monolingual performance. In evaluating pivot translation, we found that it achieved 40% performance of monolingual retrieval, which was not as good as direct translation. Overall, our results are encouraging and show promise of successful application of MLIR techniques to Web retrieval.