The models of judging character and word neighborship are presented, and the method of constructing error detecting knowledge bases which is related to these models is introduced.