The purpose of this paper is to establish the mechanism of mutual transformation between programming language and natural language through the cognitive features of programming language and natural language which enables computers to read natural language and furthermore, to think like a human brain to achieve the purpose of human-computer interaction. The method is: First, establish a data dictionary, a vocabulary table, a sentence table and a bilingual table so that natural language and programming language can establish a one-to-one, one-to-many conversion relationship. Second, according to the data dictionary and bilingual table, establish the thinking tree model, and implement the traversal of natural language (a particular mathematical problem) through programming language. Third, establish a selection mechanism so that, through programming language, the computer knows how to make different choices and responses corresponding to different natural languages. The result is: Through the exploration of the cognitive features of programming language and natural language, establish the mechanism of the mutual transformation between programming language and natural language. The significance is: through the use of two different languages, it can establish the mechanism of human-computer interaction. For the one-to-one situations, it can be automatically converted based on rules. On the other hand, for the one-to-many situation, it can realize machine learning, moreover, deep learning based on statistics. Thus, it lays the foundation for the application of artificial intelligence in the field of education.