Translation is an important index to assess students’ English learning ability. In the current English translation teaching, there are still dilemmas such as poor professionalism of curriculum, insufficient knowledge reserve of students, and single teaching methods and approaches. Therefore, this study carries out research on English translation teaching based on the deep learning model to improve the EM algorithm and the guidance of Production-Oriented Approach (POA). The experiment proves that POA can effectively mobilize students’ enthusiasm in translation learning, improve the efficiency of internalizing human knowledge into output ability, optimize teachers’ teaching effect, and significantly improve students' English translation ability and output application ability by more than 30%.