A new kind of finit-state grammer is defined as T-degree similar derivation grammer GODT having Parameter T. The structure of this grammer GCDT not only depends on the positive sample set S+

but is also controlled by Parameter T. In the process of grammatical inference

use can be made of the information provided by the negative-sample set S- to control Parameter T for modification of the inferred grammer. So the inferred grammer GCDT meets the following requirement:This learning method gives a satisfactory solution to the problem which so far has not been solved in the extent of finite-state grammer. The problem is that the inferred grammer cannot be modified by means of the negative-sample information.