PURPOSE: To improve a recognition rate by making in advance only a specified group to be a recognition object and selecting and preparing the group of a next work based on the recognition result.

CONSTITUTION: A group A0 made into a basis and subordinate cluster A1 and A2 are composed, a name made into a root is put in the group A0 made into the basis, and the blocks after the block A1 are composed of the names subordinate to respective blocks. First, the basic group A0 is recognition-processed, and when a maximum similarity does not satisfy a certain threshold, the word in the block A1 is recognition-processed, and when the maximum similarity in the block A1 satisfies the threshold, that of the block A1 is made into a recognition cluster, and a shift to a next recognition is executed. When it is not satisfied, the recognition is shifted to the next block A2. In such a way, the recognition can be executed even from the block in the course.