SISMULT- Sistema de indexação semiautomática multilíngue.

Abstract

With the outsprccd of the digital libraries and the Internet more and more electronic texts, written in several languages, become available for a wide and geographically dispersed public. This turns it's necessary to develop tools that facilitates indexing, representation and retrieval of multilingual documents. This thesis presents a method for semiautomatic construction of a multilingual thesaurus, based on the indexing of electronic documents, in order to support a adequate information retrieval, independent of the language of the documents. The method consists in extracting the terms of a document and to use an analysis of the co-occurrence of terms in order to determine its relevance. Using special unilingual dictionaries, abstract, language-independent terms are determined. Relevant concepts are represented as binary relations and, using the method of rectangular decomposition of Gammoudi, rectangles of pairs concept/document are determined and added to the existing thesaurus incrementally. Special dictionaries and an interaction with the user determines the correct contexts for ambiguous terms, further on eliminating flexions and determining the abstract concepts. A prototype has been developed which allows a continuous update of the existing thesaurus, indexing new documents, in several languages. It also supports multilingual queries and the addition of the new languages.