In this paper, we describe a crosslingual Information Retrieval System (IRS), which makes the interrogation of multilingual databases possible. Indeed, the CEA needs to be able to process an important amount of multilingual databases and documents, and so we had to adapt the IRS we use, SPIRIT (which relies on linguistical and statistical processing) to this situation. We have thus set up a crosslingual interrogation system based on the indexation of documents containing parts written in different languages and on the bilingual reformulation of the query. The latter ttries all the possible translations for every significant word of the query and the documents are used as filters in case of uncertainty or ambiguity. The answers to a query are given in the form of a list of classes of documents ranked according to their relevance. This paper describes the application of these techniques to crosslingual access to catalogs and bibliographic databases.
展开▼
