POSTECH at NTCIR-5

This paper describes methodologies for NTCIR-5 CLIR involving Korean and Japanese, and reports the official result as well as retrieval results using NTCIR-3 and NTCIR-4 data. We participated in four tasks: K-K and J-J monolingual tracks and K-J and J-K cross-lingual tracks. Unlike English, in Asian languages such as Korean and Japanese term extraction is nontrivial because of segmentation ambiguities. In this regard, we prepared multiple term representations for documents and queries, of which ranked results are merged to generate final ranking. In preliminary experiments using NTCIR-3 and NTCIR-4 data, our model showed the best performances for description queries in Korean and Japanese. In offline results using NTCIR-5 data, our methodology in Korean showed the best performance by archieving 0.5680 for description queries and 0.6159 for others.