Design and Application of English Grammar Automatic Correction System Based on Machine Learning

English grammatical errors frequently affect the clarity of information transmission, and an effective automatic error correction system is urgently needed. This paper aims to design an English grammar automatic error correction system based on machine learning. First, this paper constructs a corpus containing a large number of annotated errors and corrected samples, uses the advanced pre-trained language model BERT for feature extraction, and then fine-tunes the model through transfer learning methods to adapt to specific grammatical error correction tasks. In addition, this paper adopts a sequence-to-sequence (Seq2Seq) model architecture and combines it with an attention mechanism to enable the model to better capture contextual information and generate corresponding error correction outputs. The system can effectively identify and correct common grammatical errors, with a grammatical correction accuracy of 97.8%, improving the readability and accuracy of the text, proving its potential application value in education and writing assistance tools.