《Jacques Savoy》.pdfVIP

Proceedings of NTCIR-4, Tokyo, April 2003 - June 2004 Report on CLIR Task for the NTCIR-4 Evaluation Campaign Jacques Savoy Institut interfacultaire dinformatique, Université de Neuchâtel Pierre-à-Mazel 7, 2000 Neuchâtel, Switzerland Jacques.Savoy@unine.ch Abstract 1.1 Overview of NTCIR-4 test collection This paper describes our first participation in an evaluation campaign involving three Asian languages Table 1 displays various statistics on corpora (NTCIR-4). Our project has three objectives: 1) to made available during the fourth NTCIR evaluation compare the retrieval performances of eleven IR campaign (see also [5]). The Japanese collection is models used to carry out monolingual retrievals with the largest, the English corpus is the second largest, these languages; 2) to analyze the relative merit of with the Chinese and Korean corpus being the various freely available translation tools used to smallest. When comparing the number of distinct translate English-language topics into Chinese, bigrams per article, the Chinese documents were Japanese or Korean; and 3) to evaluate the relative usually quite large (363.4 bigrams/article), relative to performance of the various merging strategies used to the Korean (236.2) and Japanese (114.5) documents. combine separate result lists extracted from a corpus When analyzing the number of pertinent written in English, Chinese, Japanese or Korean. documents per topic,