Transitive probabilistic CLIR models
Wessel Kraaij, Franciska M. G. de Jong · University of Twente Research Information · 2004
Transitive translation could be a useful technique to enlarge the number of supported language pairs for a cross-language information retrieval (CLIR) system in a cost-effective manner. The paper describes several setups for transitive translation based on probabilistic translation models. The transitive CLIR models were evaluated on the CLEF test collection and yielded a retrieval effectiveness up to 83% of monolingual performance, which is significantly better than a baseline using the synonym operator.