This paper focuses on the differences in grammar rules and syntactic structures between Japanese passive and probable tenses and labels the corresponding sentences in the corpus according to their grammatical features. A statistical method is employed to categorize the corpus by linking dependency trees and morphological information. Differential equation modeling is used to construct a classification model, and the important features and extraction processes for distinguishing sentence structure are defined. In conclusion, the impact of language and cultural differences on the machine translation of Japanese passive sentences is examined. The results show that the significant values of manual translation and machine translation are 0.158 and 0.203, respectively, which are greater than 0.05, indicating that they conform to the normal distribution and there are obvious differences between them. The mean score of the total score of manual translation is 4.62, and the mean score of the total score of machine translation is 2.85, indicating that there is a significant effect of language and cultural differences on machine translation. This paper provides useful guidance and insights for improving the quality of Japanese translation and promoting cross-cultural communication and understanding.