With the changes in the market situation for English majors, teaching English translation in colleges and universities is also facing many challenges. This paper proposes an optimization strategy for English translation teaching methods by using machine learning technology to automatically identify English translation errors and extract text summaries. Pearson coefficient and multi-feature fusion technology are used to prejudge the correctness of English translation results, and according to the directed graph of wrong translation results, the automatic identification algorithm of English translation errors is constructed to automatically identify translation errors. The unsupervised machine learning TextRank algorithm is introduced and applied in text summary extraction, and combined with a multi-feature fusion computer system based on similarity relationships, it is improved to enhance the efficiency and quality of text extraction. Inner Mongolia Normal University set up an experimental class and a control class and applied this paper’s technology to practice English translation teaching. After the practice, the total English translation score of students in the experimental class was 85.74, which was 4.41 higher than that of the control group, showing a significant difference (P<0.05). Prior to the practice, the interest and attitude toward English translation rose from 3.42 and 2.43 to 4.32 and 4.75, while the control group’s mean values decelerated slightly. The two dimensions of teaching satisfaction, learning atmosphere, and English translation ability development were also higher than the control class by 0.71 and 0.91, indicating a statistical difference (P<0.05).