In this study, we propose a method to quantify a relation between a feature of a human motion and a word used to express its name using the subspace method. Motion name is divided into some words, and it is examined which component of a feature vector is related to each word. The proposed method is applicable to a motion database [Yukawa et al. 2000]. Using this method, 1) a search key can be automatically generated from the relationship between the features of motion data and their name when inserting human motion data into the database, and 2) we can search required human motion data from the database using a motion word as a query in the same way as is used in a search engine of a web system.