Comparison between proximity operation and dependency operation in Japanese full-text retrieval

In this paper we propose a full-text retrieval for Japanese document using dependency relation between words and evaluate it comparing to proximity operation. The proximity relation has been used as an approximation to syntactic or semantic relation because syntactic or semantic analysis with high accuracy has still been a high-cost task for a computer. We have developed the method of skeletal syntactic analysis for Japanese and apply it to the full-text retrieval. The figures of the index size, response time and accuracy of retrieval of our experiment show the effectiveness of the use of dependency relation.