CINDOR Conceptual Interlingua Document Retrieval: TREC-8 Evaluation
Text Retrieval Conference
E.M. Voorhees and D.K. Harman
The TREC-8 evaluation of the CINDOR system was based on English and French data from the cross-language retrieval track. Our objective was to continue our investigation of our conceptual interlingua approach to cross-language retrieval, specifically by measuring the contribution of conceptual retrieval over and above a baseline cross-language retrieval approach based on machine translation of queries. In both of the cross-language runs that were submitted for evaluation, corresponding to English-French and French-English retrieval, performance was measured at 75% of the equivalent monolingual searches. We noted however that absolute average precision values achieved were somewhat lower than many other systems in the cross- language track. Our hypothesis, that the underlying retrieval engine used in CINDOR was employing a simple retrieval function that was impacting performance, was confirmed through experiments with the SMART system configured with several different retrieval settings. Taken together, our TREC-8 experiments point to the value of our conceptual interlingua approach to retrieval, but indicate that our retrieval algorithm must be brought up to date so that valid comparisons may be made to other approaches used in other cross-language systems.
Ruiz, M., Diekema, A., Sheridan, P. CINDOR Conceptual Interlingua Document Retrieval: TREC-8 Evaluation. In: Proceedings of the Eighth Text Retrieval Conference (TREC-8). E.M. Voorhees and D.K. Harman (Eds.). Gaithersburg, MD: Department of Commerce, National Institute of Standards and Technology, 1999.