With the rapid development of natural language processing technology, automatic text correction plays an important role in improving text quality and communication efficiency. However, due to the complexity and diversity of language, existing text correction systems have problems such as low accuracy and insufficient generalization ability when dealing with various types of errors. To address these problems, this paper designs and implements a language automatic correction system based on convolutional neural network (CNN) and human–computer interaction. First, the system architecture and principles are designed, in which CNN extracts text features, and the human–computer interaction module performs word segmentation and annotation on the text input by the user, and provides error correction feedback to the user; then, the system performance is evaluated through experiments to measure the system performance. The experimental results show that in the public datasets CoNLL-2014, SIGHAN 2015 and self-built business text databases, the error correction accuracy of the pure CNN model is higher than that of the traditional method. The proposed method of combining CNN with human–computer interaction further improves the accuracy, and after three rounds of human–computer interaction, the error correction accuracy reaches a higher level. At the same time, the human–computer interaction feedback mechanism can effectively shorten the error correction time of different types of errors. The system has a high accuracy rate in identifying and correcting various types of language errors, which is significantly better than traditional methods and can effectively improve the quality of text.